Today marks the release of a white paper I have been working on for a long time, titled “Knowledge and Decisions in the Information Age: The Law & Economics of Regulating Misinformation on Social-Media Platforms.” In it, I attempt to outline an Austrian law & economics theory of state action under the First Amendment, and then explain why it is important to the problem of misinformation on social-media platforms.

Read the full piece here.

“If there is any fixed star in our constitutional constellation, it is that no official, high or petty, can prescribe what shall be orthodox in politics, nationalism, religion, or other matters of opinion or force citizens to confess by word or act their faith therein. If there are any circumstances which permit an exception, they do not now occur to us.” – West Virginia Board of Education v. Barnette (1943)[1]

“Our constitutional tradition stands against the idea that we need Oceania’s Ministry of Truth.” – United States v. Alvarez (2012)[2]

Introduction

In April 2022, the U.S. Department of Homeland Security (DHS) announced the creation of the Disinformation Governance Board, which would be designed to coordinate the agency’s response to the potential effects of disinformation threats.[3] Almost immediately upon its announcement, the agency was met with criticism. Congressional Republicans denounced the board as “Orwellian,”[4] and it was eventually disbanded.[5]

The DHS incident followed years of congressional hearings in which Republicans had castigated leaders of the so-called “Big Tech” firms for allegedly censoring conservatives, while Democrats had criticized those same leaders for failing to combat and remove misinformation.[6] Moreover, media outlets have reported on systematic attempts by government officials to encourage social-media companies to remove posts and users based on alleged misinformation. For example, The Intercept in 2022 reported on DHS efforts to set up backchannels with Facebook for flagging posts and misinformation.[7]

The “Twitter Files” released earlier this year by the company’s CEO Elon Musk—and subsequently reported on by journalists Barry Weiss, Matt Taibbi, and Michael Shellenberger—suggest considerable efforts by government agents to encourage Twitter to remove posts as misinformation and to bar specific users for being purveyors of misinformation.[8] What’s more, communications unveiled as part of discovery in the Missouri v. Biden case have offered further evidence a variety of government actors cajoling social-media companies to remove alleged misinformation, along with the development of a considerable infrastructure to facilitate what appears to be a joint project to identify and remove the same.[9]

With all of these details coming into public view, the question that naturally arises is what role, if any, does the government have in regulating misinformation disseminated through online platforms? The thesis of this paper is that the First Amendment forecloses government agents’ ability to regulate misinformation online, but it protects the ability of private actors—i.e., the social-media companies themselves—to regulate misinformation on their platforms as they see fit.

The primary reason for this conclusion is the state-action doctrine, which distinguishes public and private action. Public actions are subject to constitutional constraints (such as the First Amendment), while private actors are free from such regulation.[10] A further thesis of this paper is that application of the state-action doctrine to the question of misinformation on online platforms promotes the bedrock constitutional value of “protect[ing] a robust sphere of individual liberty,”[11] while also creating outlets for more speech to counteract false speech.[12]

Part I of this paper outlines a law & economics theory of state-action requirements under the First Amendment and explains its importance for the online social-media space. The right to editorial discretion and Section 230 will also be considered as part of this background law, which places the responsibility for regulating misinformation on private actors like social-media platforms. Such platforms must balance the interests of each side of their platforms to maximize value. This means, in part, setting moderation rules on misinformation that keep users engaged in order to provide increased opportunities to generate revenue from advertisers.

Part II considers various theories of state action and whether they apply to social-media platforms. It appears clear that some state-action theory—like the idea that social-media companies exercise a “traditional, exclusive public function”—are foreclosed in light of Manhattan Community Access Corp. v. Halleck. But it remains an open question whether a social-media company could be found a state actor under a coercion or collusion theory under facts that have been revealed in the Twitter Files and litigation over this question.

Part III completes the First Amendment analysis of what government agents can do to regulate misinformation on social media. The answer: not much. The U.S. Constitution forbids direct regulation of false speech simply because it is false. A more difficult question concerns how to define truth and falsity in contested areas of fact, where legal questions may run into vagueness concerns. We recommend that a better way forward is for government agents to invest in telling their own version of the facts, but where they have no authority to mandate or pressure social-media companies into regulating misinformation.

I.        A Theory of State Action and Speech Rights on Online Social-Media Platforms

Among the primary rationales for the First Amendment’s speech protections is to shield the “marketplace of ideas”:[13] in most circumstances, the best remedy for false or harmful speech is “more speech, not enforced silence.”[14] But this raises the question of why private abridgments of speech—such as those enforced by powerful online social-media platforms—should not be subject to the same First Amendment restrictions as government action.[15] After all, if the government can’t intervene in the marketplace of ideas by deciding what is true or false, then why should that privilege be held by Facebook or Google?

Here enters the state-action doctrine, which is the legal principle (discussed further below) that, in some cases, private entities may function as extensions of the state. Under this doctrine, the actions of such private actors would give rise to similar First Amendment concerns as if the state had acted on its own. It has been said that there is insufficient theorizing about the “why” of the state-action doctrine.[16] What follows is a theory of why the state-action doctrine is fundamental to protecting those private intermediaries who are best positioned to make marginal decisions about the benefits and harms of speech, including social-media companies through their moderation policies on misinformation.

Governance structures are put in place by online platforms as a response to market pressures to limit misinformation and other harmful speech. At the same time, there are also market pressures to not go too far in limiting speech.[17] The balance that must be struck by online intermediaries is delicate, and there is no reason to expect government regulators to do a better job than the marketplace in determining the optimal rules. The state-action doctrine protects a marketplace for speech governance by limiting the government’s reach into these spaces.

In order to discuss the state-action doctrine meaningfully, we must first outline its basic contours and the why identified by the Supreme Court. In Part I.A, we will provide a description of the Supreme Court’s most recent First Amendment state-action decision, Manhattan Community Access Corp. v. Halleck, where the Court both defines and defends the doctrine’s importance. We will also briefly consider how the state-action doctrine’s protection of private ordering is bolstered by the right to editorial discretion and by Section 230 of the Communications Decency Act of 1998.

We will then consider whether there are good theoretical reasons to support the First Amendment’s state-action doctrine. In Part I.B, we will apply insights from the law & economics tradition associated with the interaction of institutions and dispersed knowledge.[18] We argue that the First Amendment’s dichotomy between public and private action allows for the best use of dispersed knowledge in society by creating a marketplace for speech governance. We also argue that, by protecting this marketplace for speech governance from state action, the First Amendment creates the best institutional framework for reducing harms from misinformation.[19]

A.      The State-Action Doctrine, the Right to Editorial Discretion, and Section 230

At its most basic, the First Amendment’s state-action doctrine says that government agents may not restrict speech, whether through legislation, rules, or enforcement actions, or by putting undue burdens on speech exercised on government-owned property.[20] Such restrictions will receive varying levels of scrutiny from the courts, depending on the degree of incursion. On the other hand, the state-action doctrine means that, as a general matter, private actors may set rules for what speech they are willing to abide or promote, including rules for speech on their own property. With a few exceptions where private actors may be considered state actors,[21] these restrictions will receive no scrutiny from courts, and the government may actually help remove those who break privately set speech rules.[22]

In Halleck, the Court set out a strong defense of the state-action doctrine under the First Amendment. Justice Brett Kavanaugh, writing for the majority, defended the doctrine based on the text and purpose of the First Amendment:

Ratified in 1791, the First Amendment provides in relevant part that “Congress shall make no law … abridging the freedom of speech.” Ratified in 1868, the Fourteenth Amendment makes the First Amendment’s Free Speech Clause applicable against the States: “No State shall make or enforce any law which shall abridge the privileges or immunities of citizens of the United States; nor shall any State deprive any person of life, liberty, or property, without due process of law ….” § 1. The text and original meaning of those Amendments, as well as this Court’s longstanding precedents, establish that the Free Speech Clause prohibits only governmental abridgment of speech. The Free Speech Clause does not prohibit private abridgment of speech…

In accord with the text and structure of the Constitution, this Court’s state-action doctrine distinguishes the government from individuals and private entities. By enforcing that constitutional boundary between the governmental and the private, the state-action doctrine protects a robust sphere of individual liberty…

It is sometimes said that the bigger the government, the smaller the individual. Consistent with the text of the Constitution, the state-action doctrine enforces a critical boundary between the government and the individual, and thereby protects a robust sphere of individual liberty. Expanding the state-action doctrine beyond its traditional boundaries would expand governmental control while restricting individual liberty and private enterprise.[23]

Applying the state-action doctrine, the Court held that even the heavily regulated operation of cable companies’ public-access channels constituted private action. The Court opined that “merely hosting speech by others is not a traditional, exclusive public function and does not alone transform private entities into state actors subject to First Amendment constraints.”[24] The Court went on to explain:

If the rule were otherwise, all private property owners and private lessees who open their property for speech would be subject to First Amendment constraints and would lose the ability to exercise what they deem to be appropriate editorial discretion within that open forum. Private property owners and private lessees would face the unappetizing choice of allowing all comers or closing the platform altogether.[25]

Similarly, the Court has found that private actors have the right to editorial discretion that can’t generally be overcome by a government compelling the carriage of speech.[26] In Miami Herald v. Tornillo, the Supreme Court ruled that a right-to-reply statute for political candidates was unconstitutional because it “compel[s] editors or publishers to publish that which ‘reason tells them should not be published.’”[27] The Court found that the marketplace of ideas was still worth protecting from government-compelled speech, even in a media environment where most localities only had one (monopoly) newspaper.[28] The effect of Tornillo was to establish a general rule whereby the limits on media companies’ editorial discretion were defined not by government edict but by “the acceptance of a sufficient number of readers—and hence advertisers —to assure financial success; and, second, the journalistic integrity of its editors and publishers.”[29]

Section 230 of the Communications Decency Act supplements the First Amendment’s protections by granting “providers and users of an interactive computer service” immunity from (most) lawsuits for speech generated by other “information content providers” on their platforms.[30] The effect of this statute is far-ranging in its implications for online speech. It protects online social-media platforms from lawsuits for the third-party speech they host, as well as for the platforms’ decisions to take certain third-party speech down.[31]

As with the underlying First Amendment protections, Section 230 augments social-media companies’ ability to manage misinformation on their services. Specifically, it shields them from an unwarranted flood of litigation for failing to remove the defamatory speech of third parties when they make efforts to remove some undesirable speech from their platforms.

B.      Regulating Speech in Light of Dispersed Knowledge[32]

One of the key insights of the late Nobel laureate economist F.A. Hayek was that knowledge is dispersed.[33] In other words, no one person or centralized authority has access to all the tidbits of knowledge possessed by countless individuals spread out through society. Even the most intelligent among us have but a little bit more knowledge than the least intelligent. Thus, the economic problem facing society is not how to allocate “given” resources, but how to “secure the best use of resources known to any of the members of society, for ends whose relative importance only these individuals know.”[34]

This is particularly important when considering the issue of regulating alleged misinformation. As noted above, the First Amendment is premised on the idea that a marketplace of ideas will lead to the best information eventually winning out, with false ideas pushed aside by true ones.[35] Much like the economic problem, there are few, if any, given answers that are true for all time when it comes to opinions or theories in science, the arts, or any other area of knowledge. Thus, the question is: how do we establish a system that promotes the generation and adoption of knowledge, recognizing there will be “market failures” (and possibly, corresponding “government failures”) along the way?

Like virtually any other human activity, there are benefits and costs to speech. It is ultimately subjective individual preference that determines how to manage those tradeoffs. Although the First Amendment protects speech from governmental regulation, that does not mean that all speech is acceptable or must be tolerated. As noted above, U.S. law places the power to decide what speech to allow in the public square firmly into the hands of the people. The people’s preferences are expressed individually and collectively through their participation in online platforms, news media, local organizations, and other fora, and it via that process that society arrives at workable solutions to such questions.

Very few people believe that all speech protected by the First Amendment should be without consequence. Just as very few people, if pressed, would really believe that it is, generally speaking, a wise idea to vest the power to determine what is true or false in a vast governmental bureaucracy. Instead, proposals for government regulation of misinformation generally are offered as an expedient to effect short-term political goals that are perceived to be desirable. But given the dispersed nature of knowledge and given that very few “facts” are set in stone for all time,[36] such proposals threaten to undermine the very process through which new knowledge is discovered and disseminated.

Moreover, such proposals completely fail to account for how “bad” speech has, in fact, long been regulated via informal means, or what one might call “private ordering.” In this sense, property rights have long played a crucial role in determining the speech rules of any given space. If a man were to come into another man’s house and start calling his wife racial epithets, he would not only have the right to ask that person to leave but could exercise his right as a property owner to eject the trespasser—if necessary, calling the police to assist him. One similarly could not expect to go to a restaurant and yell at the top of her lungs about political issues and expect the venue—even those designated as “common carriers” or places of public accommodation—to allow her to continue.[37] A Christian congregation may in most circumstances be extremely solicitous of outsiders with whom they want to share their message, but they would likewise be well within their rights to prevent individuals from preaching about Buddhism or Islam within their walls.

In each of these examples, the individual or organization is entitled to eject individuals on the basis of their offensive (or misinformed) speech with no cognizable constitutional complaint about the violation of rights to free speech. The nature of what is deemed offensive is obviously context- and listener-dependent, but in each example, the proprietors of the relevant space are able to set and enforce appropriate speech rules. By contrast, a centralized authority would, by its nature, be forced to rely on far more generalized rules. As the economist Thomas Sowell once put it:

The fact that different costs and benefits must be balanced does not in itself imply who must balance them?or even that there must be a single balance for all, or a unitary viewpoint (one “we”) from which the issue is categorically resolved.[38]

When it comes to speech, the balance that must be struck is between one individual’s desire for an audience and that prospective audience’s willingness to listen. Asking government to make categorical decisions for all of society is substituting centralized evaluation of the costs and benefits of access to communications for the individual decisions of many actors. Rather than incremental decisions regarding how and under what terms individuals may relate to one another—which can evolve over time in response to changes in what individuals find acceptable—governments can only hand down categorical guidelines: “you must allow a, b, and c speech” or “you must not allow z, y, and z speech.”

It is therefore a fraught proposition to suggest that government could have both a better understanding of what is true and false, and superior incentives to disseminate the truth, than the millions of individuals who make up society.[39] Indeed, it is a fundamental aspect of both the First Amendment’s Establishment Clause[40] and of free-speech jurisprudence[41] that the government is in no position to act as an arbiter of what is true or false.

Thus, as much as the First Amendment protects a marketplace of ideas, by excluding the government as a truth arbiter, it also protects a marketplace for speech governance. Private actors can set the rules for speech on their own property, including what is considered true or false, with minimal interference from the government. And as the Court put it in Halleck, opening one’s property for the speech of third parties need not make the space take all-comers.[42]

This is particularly relevant in the social-media sphere. Social-media companies must resolve social-cost problems among their users.[43] In his famous work “The Problem of Social Cost,” the economist Ronald Coase argued that the traditional approach to regulating externalities was wrong, because it failed to apprehend the reciprocal nature of harms.[44] For example, the noise from a factory is a potential cost to the doctor next door who consequently can’t use his office to conduct certain testing, and simultaneously the doctor moving his office next door is a potential cost to the factory’s ability to use its equipment. In a world of well-defined property rights and low transaction costs, the initial allocation of a right would not matter, because the parties could bargain to overcome the harm in a beneficial manner—i.e., the factory could pay the doctor for lost income or to set up sound-proof walls, or the doctor could pay the factory to reduce the sound of its machines.[45] Similarly, on social media, misinformation and other speech that some users find offensive may be inoffensive or even patently true to other users. There is a reciprocal nature to the harms of offensive speech, much as with other forms of nuisance. But unlike the situation of the factory owner and the doctor, social-media users use the property of social-media companies, who must balance these varied interests to maximize the platform’s value.

Social-media companies are what economists call “multi-sided” platforms.[46] They are profit seeking, to be sure, but the way they generate profits is by acting as intermediaries between users and advertisers. If they fail to serve their users well, those users will abandon the platform. Without users, advertisers would have no interest in buying ads. And without advertisers, there is no profit to be made. Social-media companies thus need to maximize the value of their platform by setting rules that keep users sufficiently engaged that there are advertisers who will pay to reach them.

In the cases of Facebook, Twitter, and YouTube, the platforms have set content-moderation standards that restrict many kinds of speech, including misinformation. [47] In some cases, these policies are viewed negatively by some users, particularly given that the First Amendment would foreclose the government from regulating those same types of content. But social-media companies’ ability to set and enforce moderation policies could actually be speech-enhancing. Because social-media companies are motivated to maximize the value of their platforms, for any given policy that gives rise to enforcement actions that leave some users disgruntled, there are likely to be an even greater number of users who agree with the policy. Moderation policies end up being speech-enhancing when they promote more speech overall, as the proliferation of harmful speech may push potential users away from the platforms.

Currently, all social-media companies rely on an advertising-driven revenue model. As a result, their primary goal is to maximize user engagement. As we have recently seen, this can lead to situations where advertisers threaten to pull ads if they don’t like the platform’s speech-governance decisions. After Elon Musk began restoring the accounts of Twitter users who had been banned for what the company’s prior leadership believed was promoting hate speech and misinformation, major advertisers left the platform.[48] A different business model (about which Musk has been hinting for some time[49]) might generate different incentives for what speech to allow and disallow. There would, however, still be a need for any platform to allow some speech and not other speech, in line with the expectations of its user base and advertisers. The bottom line is that the motive to maximize profits and the tendency of markets to aggregate information leaves the platforms themselves best positioned to make these incremental decisions about their users’ preferences, in response to the feedback mechanism of consumer demand.

Moreover, there is a fundamental difference between private action and state action, as alluded to by the Court in Halleck: one is voluntary, and the other based on coercion. If Facebook or Twitter suspends a user for violating community rules, that decision terminates a voluntary association. When the government removes someone from a public forum for expressing legal speech, its censorship and use of coercion are inextricably intertwined. The state-action doctrine empowers courts to police this distinction because the threats to liberty are much greater when one party in a dispute over the content of a particular expression is also empowered to impose its will with the use of force.

Imagine instead that courts were to decide that they, in fact, were best situated to balance private interests in speech against other interests, or even among speech interests. There are obvious limitations on courts’ access to knowledge that couldn’t be easily overcome through the processes of adjudication, which depend on the slow development of articulable facts and categorical reasoning over a lengthy period of time and an iterative series of cases. Private actors, on the other hand, can act relatively quickly and incrementally in response to ever-changing consumer demand in the marketplace. As Sowell put it:

The courts’ role as watchdogs patrolling the boundaries of governmental power is essential in order that others may be secure and free on the other side of those boundaries. But what makes watchdogs valuable is precisely their ability to distinguish those people who are to be kept at bay and those who are to be left alone. A watchdog who could not make that distinction would not be a watchdog at all, but simply a general menace.

The voluntariness of many actions—i.e., personal freedom—is valued by many simply for its own sake. In addition, however, voluntary decision-making processes have many advantages which are lost when courts attempt to prescribe results rather than define decision-making boundaries.[50]

The First Amendment’s complementary right of editorial discretion also protects the right of publishers, platforms, and other speakers to be free from an obligation to carry or transmit government-compelled speech.[51] In other words, not only is private regulation of speech not state action, but as a general matter, private regulation of speech is protected by the First Amendment from government action. The limits on editorial discretion are marketplace pressures, such as user demand and advertiser support, and social mores about what is acceptable to be published.[52]

There is no reason to think that social-media companies today are in a different position than was the newspaper in Tornillo.[53] These companies must determine what, how, and where content is presented within their platform. While this right of editorial discretion protects social-media companies’ moderation decisions, its benefits accrue to society at-large, who get to use those platforms to interact with people from around the world and to thereby grow the “marketplace of ideas.”

Moreover, Section 230 amplifies online platforms’ ability to make editorial decisions by immunizing most of their choices about third-party content. In fact, it is interesting to note that the heading for Section 230 is “Protection for private blocking and screening of offensive material.”[54] In other words, Section 230 is meant, along with the First Amendment, to establish a market for speech governance free from governmental interference.

Social-media companies’ abilities to differentiate themselves based on functionality and moderation policies are important aspects of competition among them.[55] How each platform is used may differ depending on those factors. In fact, many consumers use multiple social-media platforms throughout the day for different purposes.[56] Market competition, not government power, has enabled internet users to have more avenues than ever to get their message out.[57]

If social-media users and advertisers demand less of the kinds of content commonly considered to be misinformation, platforms will do their best to weed those things out. Platforms won’t always get these determinations right, but it is by no means clear that centralizing decisions about misinformation by putting them in the hands of government officials would promote the societal interest in determining the truth.

It is true that content-moderation policies make it more difficult for speakers to communicate some messages, but that is precisely why they exist. There is a subset of protected speech to which many users do not wish to be subject, including at least some perceived misinformation. Moreover, speakers have no inherent right to an audience on a social-media platform. There are always alternative means to debate the contested issues of the day, even if it may be more costly to access the desired audience.

In sum, the First Amendment’s state-action doctrine assures us that government may not make the decision about what is true or false, or to restrict a citizen’s ability to reach an audience with ideas. Governments do, however, protect social-media companies’ rights to exercise editorial discretion on their own property, including their right to make decisions about regulating potential misinformation. This puts the decisions in the hands of the entities best placed to balance the societal demands for online speech and limits on misinformation. In other words, the state-action doctrine protects the marketplace of ideas.

II.      Are Online Platforms State Actors?

As the law currently stands, the First Amendment grants online platforms the right to exercise their own editorial discretion, free from government intervention. By contrast, if government agents pressure or coerce platforms into declaring certain speech misinformation, or to remove certain users, a key driver of the marketplace of ideas—the action of differentiated actors experimenting with differing speech policies—will be lost.[58]

Today’s public debate is not actually centered on a binary choice between purely private moderation and legislatively enacted statutes to literally define what is true and what is false. Instead, the prevailing concerns relate to the circumstances under which some government activity—such as chastising private actors for behaving badly, or informing those actors about known threats—might transform online platforms’ moderation policies into de facto state actions. That is, at what point do private moderation decisions constitute state action? To this end, we will now consider sets of facts under which online platforms could be considered state actors for the purposes of the First Amendment.

In Halleck, the Supreme Court laid out three exceptions to the general rule that private actors are not state actors:

Under this Court’s cases, a private entity can qualify as a state actor in a few limited circumstances—including, for example, (i) when the private entity performs a traditional, exclusive public function; (ii) when the government compels the private entity to take a particular action; or (iii) when the government acts jointly with the private entity.[59]

Below, we will consider each of these exceptions, as applied to online social-media platforms. Part II.A will make the case that Halleck decisively forecloses the theory that social-media platforms perform a “traditional, exclusive public function,” as has been found by many federal courts. Part II.B will consider whether government agents have coerced or encouraged platforms to make specific enforcement decisions on misinformation in ways that would transform their moderation actions into state action. Part II.C will look at whether the social-media companies have essentially colluded with government actors, through either joint action or in a relationship sufficiently intertwined as to be symbiotic.

A.      ‘Traditional, Exclusive Public Function’

The classic case that illustrates the traditional, exclusive public function test is Marsh v. Alabama.[60] There, the Supreme Court found that a company town, while private, was a state actor for the purposes of the First Amendment. At issue was whether the company town could prevent a Jehovah’s Witness from passing out literature on the town’s sidewalks. The Court noted that “[o]wnership does not always mean absolute dominion. The more an owner, for his advantage, opens up his property for use by the public in general, the more do his rights become circumscribed by the statutory and constitutional rights of those who use it.”[61] The Court then situated the question as one where it was being asked to balance property rights with First Amendment rights. Within that framing, it found that the First Amendment’s protections should be in the “preferred position.”[62]

Despite nothing in Marsh suggesting a limitation to company towns or the traditional, exclusive public function test, future courts eventually cabined it. But there was a time when it looked like the Court would expand this reasoning to other private actors who were certainly not engaged in a traditional, exclusive public function. A trio of cases involving shopping malls eventually ironed this out.

First, in Food Employees v. Logan Valley Plaza,[63] the Court—noting the “functional equivalence” of the business block in Marsh and the shopping center[64] —found that the mall could not restrict the peaceful picketing of a grocery store by a local food-workers union.[65]

But then, the Court seemingly cabined-in both Logan Valley and Marsh just a few years later in Lloyd Corp. v. Tanner.[66] Noting the “economic anomaly” that was company towns, the Court said Marsh “simply held that where private interests were substituting for and performing the customary functions of government, First Amendment freedoms could not be denied where exercised in the customary manner on the town’s sidewalks and streets.”[67] Moreover, the Court found that Logan Valley applied “only in a context where the First Amendment activity was related to the shopping center’s operations.”[68] The general rule, according to the Court, was that private actors had the right to restrict access to property for the purpose of exercising free-speech rights.[69] Importantly, “property does not lose its private character merely because the public is generally invited to use it for designated purposes.”[70] Since the mall did not dedicate any part of its shopping center to public use in a way that would entitle the protestors to use it, the Court allowed it to restrict hand billing by Vietnam protestors within the mall.[71]

Then, in Hudgens v. NLRB,[72] the Court went a step further and reversed Logan Valley and severely cabined-in Marsh. Now, the general rule was that “the constitutional guarantee of free speech is a guarantee only against abridgment by government, federal or state.”[73] Marsh is now a narrow exception, limited to situations where private property has taken on all attributes of a town.[74] The Court also found that the reasoning—if not the holding—of Tanner had already reversed Logan Valley.[75] The Court concluded bluntly that “under the present state of the law the constitutional guarantee of free expression has no part to play in a case such as this.”[76] In other words, private actors, even those that open themselves up to the public, are not subject to the First Amendment. Following Hudgens, the Court would further limit the public-function test to “the exercise by a private entity of powers traditionally exclusively reserved to the State.”[77] Thus, the “traditional, exclusive public function” test.

Despite this history, recent litigants against online social-media platforms have argued, often citing Marsh, that these platforms are the equivalent of public parks or other public forums for speech.[78] On top of that, the Supreme Court itself has described social-media platforms as the “modern public square.”[79] The Court emphasized the importance of online platforms because they:

allow[] users to gain access to information and communicate with one another about it on any subject that might come to mind… [give] access to what for many are the principal sources for knowing current events, checking ads for employment, speaking and listening in the modern public square, and otherwise exploring the vast realms of human thought and knowledge. These websites can provide perhaps the most powerful mechanisms available to a private citizen to make his or her voice heard. They allow a person with an Internet connection to “become a town crier with a voice that resonates farther than it could from any soapbox.”[80]

Seizing upon this language, many litigants have argued that online social-media platforms are public forums for First Amendment purposes. To date, all have failed in federal court under this theory,[81] and the Supreme Court officially foreclosed it in Halleck.

In Halleck, the Court considered whether a public-access channel operated by a cable provider was a government actor for purposes of the First Amendment under the traditional, exclusive public function test. Summarizing the caselaw, the Court said the test required more than just a finding that the government at some point exercised that function, or that the function serves the public good. Instead, the government must have “traditionally and exclusively performed the function.”[82]

The Court then found that operating as a public forum for speech is not a function traditionally and exclusively performed by the government. On the contrary, a private actor that provides a forum for speech normally retains “editorial discretion over the speech and speakers in the forum”[83] because “[it] is not an activity that only governmental entities have traditionally performed.”[84] The Court reasoned that:

If the rule were otherwise, all private property owners and private lessees who open their property for speech would be subject to First Amendment constraints and would lose the ability to exercise what they deem to be appropriate editorial discretion within that open forum. Private property owners and private lessees would face the unappetizing choice of allowing all comers or closing the platform altogether.[85]

If the applicability of Halleck to the question of whether online social-media platforms are state actors under the “traditional, exclusive public function” test isn’t already clear, there have been appellate courts who have squarely addressed the question. In Prager University v. Google, LLC,[86] the 9th U.S. Circuit Court of Appeals took on the question of whether social-media platforms are state actors subject to First Amendment. Prager relied primarily upon Marsh and Google’s representations that YouTube is a “public forum” to argue that YouTube is a state actor under the traditional, public function test.[87] Citing primarily Halleck, along with a healthy dose of both Hudgens and Tanner, the 9th Circuit rejected this argument, for the reasons noted above. [88] YouTube was not a state actor just because it opened itself up to the public as a forum for free speech.

In sum, there is no basis for arguing that online social-media platforms fit into the narrow Marsh exception to the general rule that private actors can use their own editorial discretion over own their digital property to set their own rules for speech, including misinformation policies.

That this exception to the general private/state action dichotomy has been limited as applied to social-media platforms is consistent with the reasoning laid out above on the law & economics of the doctrine. Applying the Marsh theory to social-media companies would make all of their moderation decisions subject to First Amendment analysis. As will be discussed more below in Part III.A, this would severely limit the platforms’ ability to do anything at all with regard to online misinformation, since government actors can do very little to regulate such speech consistent with the First Amendment.

The inapplicability of the Marsh theory of state action means that a robust sphere of individual liberty will be protected. Social-media companies will be able to engage in a vibrant “market for speech governance” with respect to misinformation, responding to the perceived demands of users and advertisers and balancing those interests in a way that maximizes the value of their platforms in the presence of market competition.

B.      Government Compulsion or Encouragement

In light of the revelations highlighted in the introduction of this paper from The Intercept, the “Twitter Files,” and subsequent litigation in Missouri v. Biden,[89] the more salient theory of state action is that online social-media companies were either compelled by or colluded in joint action with the federal government to censor speech under their misinformation policies. This section will consider the government compulsion or encouragement theory and Part II.C below will consider the joint action/entwinement theory.

At a high level, the government may not coerce or encourage private actors to do what it may itself not do constitutionally.[90] But state action can be found for a private decision under this theory “only when it has exercised coercive power or has provided such significant encouragement, either overt or cover, that the choice must in law be deemed to be that of the State.”[91] But “[m]ere approval of or acquiescence in the initiatives of a private party is not sufficient to justify holding the State responsible” for private actions.[92] While each case is very fact-specific,[93] courts have developed several tests to determine when government compulsion or encouragement would transform a private actor into a state actor for constitutional purposes.

For instance, in Bantam Books v. Sullivan,[94] the Court considered whether letters sent by a legislatively created commission to book publishers declaring certain books and magazines objectionable for sale or distribution was sufficient to transform into state action the publishers’ subsequent decision not to publish further copies of the listed publications. The commission had no legal power to apply formal legal sanctions and there were no bans or seizures of books.[95] In fact, the book distributors were technically “free” to ignore the commission’s notices.[96] Nonetheless, the Court found “the Commission deliberately set about to achieve the suppression of publications deemed ‘objectionable’ and succeeded in its aim.”[97] Particularly important to the Court was that the notices could be seen as a threat to refer them for prosecution, regardless how the commission styled them. As the Court stated:

People do not lightly disregard public officers’ thinly veiled threats to institute criminal proceedings against them if they do not come around, and [the distributor’s] reaction, according to uncontroverted testimony, was no exception to this general rule. The Commission’s notices, phrased virtually as orders, reasonably understood to be such by the distributor, invariably followed up by police visitations, in fact stopped the circulation of the listed publications ex proprio vigore. It would be naive to credit the State’s assertion that these blacklists are in the nature of mere legal advice, when they plainly serve as instruments of regulation…[98]

Similarly, in Carlin Communications v. Mountain States Telephone Co.,[99] the 9th U.S. Circuit Court of Appeals found it was state action when a deputy county attorney threatened prosecution of a regional telephone company for carrying an adult-entertainment messaging service.[100] “With this threat, Arizona ‘exercised coercive power’ over Mountain Bell and thereby converted its otherwise private conduct into state action…”[101] The court did not find it relevant whether or not the motivating reason for the removal was the threat of prosecution or the telephone company’s independent decision.[102]

In a more recent case dealing with Backpage.com, the 7th U.S. Circuit Court of Appeals found a sheriff’s campaign to shut down the site by cutting off payment processing for ads from Visa and Mastercard was impermissible under the First Amendment.[103] There, the sheriff sent a letter to the credit-card companies asking them to “cease and desist” from processing payment for advertisements on Backpage.com and for “contact information” for someone within the companies he could work with.[104] The court spent considerable time distinguishing between “attempts to convince and attempts to coerce,”[105] coming to the conclusion that “Sheriff Dart is not permitted to issue and publicize dire threats against credit card companies that process payments made through Backpage’s website, including threats of prosecution (albeit not by him, but by other enforcement agencies that he urges to proceed against them), in an effort to throttle Backpage.”[106] The court also noted “a threat is actionable and thus can be enjoined even if it turns out to be empty—the victim ignores it, and the threatener folds his tent.”[107]

In sum, the focus under the coercion or encouragement theory is on what the state objectively did and not on the subjective understanding of the private actor. In other words, the question is whether the state action is reasonably understood as coercing or encouraging private action, not whether the private actor was actually responding to it.

To date, several federal courts have dismissed claims that social-media companies are state actors under the compulsion/encouragement theory, often distinguishing the above cases on the grounds that the facts did not establish a true threat, or were not sufficiently connected to the enforcement action again the plaintiff.

For instance, in O’Handley v. Weber,[108] the 9th U.S. Circuit Court of Appeals dealt directly with the question of the coercion theory in the context of social-media companies moderating misinformation, allegedly at the behest of California’s Office of Elections Cybersecurity (OEC). The OEC flagged allegedly misleading posts on Facebook and Twitter and the social-media companies removed most of those flagged posts.[109] First, the court found there was no threats from the OEC like those in Carlin, nor any incentive offered to take the posts down.[110]  The court then distinguished between “attempts to convince and attempts to coerce,”[111] noting that “[a] private party can find the government’s stated reasons for making a request persuasive, just as it can be moved by any other speaker’s message. The First Amendment does not interfere with this communication so long as the intermediary is free to disagree with the government and to make its own independent judgment about whether to comply with the government’s request.”[112] The court concluded that the OEC did not pressure Twitter to take any particular action against the plaintiff, but went even further by emphasizing that, even if their actions could be seen as a specific request to remove his post, Twitter’s compliance was “purely optional.”[113] In other words, if there is no threat in a government actor’s request to take down content, then it is not impermissible coercion or encouragement.

In Hart v. Facebook,[114] the plaintiff argued that the federal government defendants had—through threats of removing Section 230 immunity and antitrust investigations, as well as comments by President Joe Biden stating that social-media companies were “killing people” by not policing misinformation about COVID-19—coerced Facebook and Twitter into removing his posts.[115] The plaintiff also pointed to recommendations from Biden and an advisory from Surgeon General Vivek Murthy as further evidence of coercion or encouragement. The court rejected this evidence, stating that “the government’s vague recommendations and advisory opinions are not coercion. Nor can coercion be inferred from President Biden’s comment that social media companies are ‘killing people’… A President’s one-time statement about an industry does not convert into state action all later decisions by actors in that industry that are vaguely in line with the President’s preferences.”[116] But even more importantly, the court found that there was no connection between the allegations of coercion and the removal of his particular posts: “Hart has not alleged any connection between any (threat of) agency investigation and Facebook and Twitter’s decisions… even if Hart had plausibly pleaded that the Federal Defendants exercised coercive power over the companies’ misinformation policies, he still fails to specifically allege that they coerced action as to him.”[117]

Other First Amendment cases against social-media companies alleging coercion or encouragement from state actors have been dismissed for reasons similar to those in Hart.[118] In Missouri et al. v. Biden, et al.,[119] the U.S. District Court for the Western District of Louisiana became the first court to find social-media companies could be state actors for purposes of the First Amendment due to a coercion or encouragement theory. After surveying (most of the same) cases as above, the court found that:

Here, Plaintiffs have clearly alleged that Defendants attempted to convince social-media companies to censor certain viewpoints. For example, Plaintiffs allege that Psaki demanded the censorship of the “Disinformation Dozen” and publicly demanded faster censorship of “harmful posts” on Facebook. Further, the Complaint alleges threats, some thinly veiled and some blatant, made by Defendants in an attempt to effectuate its censorship program. One such alleged threat is that the Surgeon General issued a formal “Request for Information” to social-media platforms as an implied threat of future regulation to pressure them to increase censorship. Another alleged threat is the DHS’s publishing of repeated terrorism advisory bulletins indicating that “misinformation” and “disinformation” on social-media platforms are “domestic terror threats.” While not a direct threat, equating failure to comply with censorship demands as enabling acts of domestic terrorism through repeated official advisory bulletins is certainly an action social-media companies would not lightly disregard. Moreover, the Complaint contains over 100 paragraphs of allegations detailing “significant encouragement” in private (i.e., “covert”) communications between Defendants and social-media platforms.

The Complaint further alleges threats that far exceed, in both number and coercive power, the threats at issue in the above-mentioned cases. Specifically, Plaintiffs allege and link threats of official government action in the form of threats of antitrust legislation and/or enforcement and calls to amend or repeal Section 230 of the CDA with calls for more aggressive censorship and suppression of speakers and viewpoints that government officials disfavor. The Complaint even alleges, almost directly on point with the threats in Carlin and Backpage, that President Biden threatened civil liability and criminal prosecution against Mark Zuckerberg if Facebook did not increase censorship of political speech. The Court finds that the Complaint alleges significant encouragement and coercion that converts the otherwise private conduct of censorship on social-media platforms into state action, and is unpersuaded by Defendants’ arguments to the contrary.[120]

There is obvious tension between Missouri v. Biden and the O’Handley and Hart opinions. As noted above, the Missouri v. Biden court did attempt to incorporate O’Handley into its opinion. That court tried to distinguish O’Handley on the grounds that the OEC’s conduct at issue was a mere advisory, whereas the federal defendants in Missouri v. Biden made threats against the plaintiffs.[121]

It is perhaps plausible that Hart can also be read as consistent with Missouri v. Biden, in the sense that while Hart failed to allege sufficient facts of coercion/encouragement or a connection with his specific removal, the plaintiffs in Missouri v. Biden did. Nonetheless, the Missouri v. Biden court accepted many factual arguments that were rejected in Hart, such as those about the relevance of certain statements made by President Biden and his press secretary; threats to revoke Section 230 liability protections; and threats to start antitrust proceedings. Perhaps the difference is that the factual allegations in Missouri v. Biden were substantially longer and more detailed than those in Hart. And while the Missouri v. Biden court did not address it in its First Amendment section, they did note that the social-media companies’ censorship actions generated sufficient injury-in-fact to the plaintiffs to establish standing.[122] In other words, it could just be that what makes the difference is the better factual pleading in Missouri v. Biden, due to more available revelations of government coercion and encouragement.[123]

On the other hand, there may be value to cabining Missouri v. Biden with some of the criteria in O’Handley and Hart. For instance, there could be value in the government having the ability to share information with social-media companies and make requests to review certain posts and accounts that may purvey misinformation. O’Handley emphasizes that there is a difference between convincing and coercing. This is not only important for dealing with online misinformation, but with things like terrorist activity on the platforms. Insofar as Missouri v. Biden is too lenient in allowing cases to go forward, this may be a fruitful distinction for courts to clarify.[124]

Similarly, the requirement in Hart that a specific moderation decision be connected to a particular government action is very important to limit the universe of activity subject to First Amendment analysis. The Missouri v. Biden court didn’t deal sufficiently with whether the allegations of coercion and encouragement were connected to the plaintiffs’ content and accounts being censored. As Missouri v. Biden reaches the merits stage of the litigation, the court will also need to clarify the evidence needed to infer state action, assuming there is no explicit admission of direction by state actors.[125]

Under the law & economics theory laid out in Part I, the coercion or encouragement exception to the strong private/state action distinction is particularly important. The benefits of private social-media companies using their editorial judgment to remove misinformation in response to user and advertiser demand is significantly reduced when the government coerces, encourages, or otherwise induces moderation decisions. In such cases, the government is essentially engaged in covert regulation by deciding for private actors what is true and what is false. This is inconsistent with a “marketplace of ideas” or the “marketplace for speech governance” that the First Amendment’s state-action doctrine protects.

There is value, however, to limiting the Missouri v. Biden holding to ensure that not all requests by government agents automatically transform moderation decisions into state action, and in connecting coercion or encouragement to particular allegations of censorship. Government actors, as much as private actors, should be able to alert social-media companies to the presence of misinformation and even persuade social-media companies to act in certain cases, so long as that communication doesn’t amount to a threat. This is consistent with a “marketplace for speech governance.” Moreover, social-media companies shouldn’t be considered state actors for all moderation decisions, or even all moderation decisions regarding misinformation, due to government coercion or encouragement in general. Without a nexus between the coercion or encouragement and a particular moderation decision, social-media companies would lose the ability to use their editorial judgment on a wide variety of issues in response to market demand, to the detriment of their users and advertisers.

C.      Joint Action or Symbiotic Relationship

There is also state action for the purposes of the First Amendment when the government acts jointly with a private actor,[126] when there is a “symbiotic relationship” between the government and a private actor,[127] or when there is “inextricable entwinement” between a private actor and the government.[128] None of these theories is necessarily distinct,[129] and it is probably easier to define them through examples.[130]

In Lugar v. Edmonson Oil Co., the plaintiff, an operator of a truck stop, was indebted to his supplier.[131] The defendant was a creditor who used a state law in Virginia to get a prejudgment attachment to the truck-stop operator’s property, which was then executed by the county sheriff.[132] A hearing was held 34 days later, pursuant to the relevant statute.[133] The levy at-issue was dismissed because the creditor failed to satisfy the statute. The plaintiff then brought a Section 1983 claim against the defendant on grounds that it had violated the plaintiff’s Due Process rights by taking his property without first providing him with a hearing. The Supreme Court took the case to clarify how the state-action doctrine applied in such matters. The Court, citing previous cases, stated that:

Private persons, jointly engaged with state officials in the prohibited action, are acting “under color” of law for purposes of the statute. To act “under color” of law does not require that the accused be an officer of the State. It is enough that he is a willful participant in joint activity with the State or its agents.[134]

The Court also noted that “we have consistently held that a private party’s joint participation with state officials in the seizure of disputed property is sufficient to characterize that party as a ‘state actor.’”[135] Accordingly, the Court found that the defendant’s use of the prejudgment statute was state action that violated Due Process.[136]

In Burton v. Wilmington Parking Authority,[137] the Court heard a racial-discrimination case in which the question was whether state action was involved when a restaurant refused to serve black customers in a space leased from a publicly owned building attached to a public parking garage.[138] The Court determined that it was state action, noting that “[i]t cannot be doubted that the peculiar relationship of the restaurant to the parking facility in which it is located confers on each an incidental variety of mutual benefits… Addition of all these activities, obligations and responsibilities of the Authority, the benefits mutually conferred, together with the obvious fact that the restaurant is operated as an integral part of a public building devoted to a public parking service, indicates that degree of state participation and involvement in discriminatory action which it was the design of the Fourteenth Amendment to condemn.”[139] While Court didn’t itself call this theory the “symbiotic relationship” test in Burton, later Court opinions did exactly that.[140]

Brentwood Academy v. Tennessee Secondary School Athletic Association arose concerned a dispute between a private Christian school and the statewide athletics association governing interscholastic sports over a series of punishments for alleged “undue influence” in recruiting athletes.[141] The central issue was whether the athletic association was a state actor. The Court analyzed whether state actors were so “entwined” with the private actors in the association to make the resulting action state action.[142] After reviewing the record, the Court noted that 84% of the members of the athletic association were public schools and the association’s rules were made by representatives from those schools.[143] The Court concluded that the “entwinement down from the State Board is therefore unmistakable, just as the entwinement up from the member public schools is overwhelming. Entwinement will support a conclusion that an ostensibly private organization ought to be charged with a public character and judged by constitutional standards; entwinement to the degree shown here requires it.”[144]

Other cases have also considered circumstances in which government regulation, combined with other government actions, can create a situation where private action is considered that of the government. In Skinner v. Railway Labor Executives Association,[145] the Court considered a situation where private railroads engaged in drug testing of employees, pursuant to a federal regulation that authorized them to adopt a policy of drug testing and preempted state laws restricting testing.[146] The Court stated that “[t]he fact that the Government has not compelled a private party to perform a search does not, by itself, establish that the search is a private one. Here, specific features of the regulations combine to convince us that the Government did more than adopt a passive position toward the underlying private conduct.”[147] The Court found the preemption of state law particularly important, finding “[t]he Government has removed all legal barriers to the testing authorized by Subpart D and indeed has made plain not only its strong preference for testing, but also its desire to share the fruits of such intrusions.”[148]

Each of these theories has been pursued by litigants who have had social-media posts or accounts removed by online platforms due to alleged misinformation, including in the O’Handley and Hart cases discussed earlier.

For instance, in O’Handley, the 9th Circuit rejected that Twitter was a state actor under the joint-action test. The court stated there were two ways to prove joint action: either by a conspiracy theory that required a “meeting of the minds” to violate constitutional rights, or by a “willful participant” theory that requires “a high degree of cooperation between private parties and state officials.”[149] The court rejected the conspiracy theory, stating there was no meeting of the minds to violate constitutional rights because Twitter had its own independent interest in “not allowing users to leverage its platform to mislead voters.”[150] The court also rejected the willful-participant theory because Twitter was free to consider and reject flags made by the OEC in the Partner Support Portal under its own understanding of its policy on misinformation.[151] The court analogized the case to Mathis v. Pac. Gas & Elec. Co.,[152] finding this “closely resembles the ‘consultation and information sharing’ that we held did not rise to the level of joint action.”[153] The court concluded that “this was an arm’s-length relationship, and Twitter never took its hands off the wheel.”[154]

Similarly, in Hart, the U.S. District Court for the Northern District of California rejected the joint action theory as applied to Twitter and Facebook. The court found that much of the complained-of conduct by Facebook predated the communications with the federal defendants about misinformation, making it unlikely that there was a “meeting of the minds” to deprive the plaintiff of his constitutional rights.[155] The court also found “the Federal Defendants’ statements… far too vague and precatory to suggest joint action,” adding that recommendations and advisories are both vague and unenforceable.[156] Other courts followed similar reasoning in rejecting First Amendment claims against social-media companies.[157]

Finally, in Children’s Health Defense v. Facebook,[158] the court considered the argument of whether Section 230, much like the regulation at issue in Skinner, could make Facebook into a joint actor with the state when it removes misinformation. The U.S. District Court for the Northern District of California distinguished Skinner, citing a previous case finding “[u]nlike the regulations in Skinner, Section 230 does not require private entities to do anything, nor does it give the government a right to supervise or obtain information about private activity.”[159]

For the first time, a federal district court found state action under the joint action or entwinement theory in Missouri v. Biden. The court found that:

Here, Plaintiffs have plausibly alleged joint action, entwinement, and/or that specific features of Defendants’ actions combined to create state action. For example, the Complaint alleges that “[o]nce in control of the Executive Branch, Defendants promptly capitalized on these threats by pressuring, cajoling, and openly colluding with social-media companies to actively suppress particular disfavored speakers and viewpoints on social media.” Specifically, Plaintiffs allege that Dr. Fauci, other CDC officials, officials of the Census Bureau, CISA, officials at HHS, the state department, and members of the FBI actively and directly coordinated with social-media companies to push, flag, and encourage censorship of posts the Government deemed “Mis, Dis, or Malinformation.”[160]

The court also distinguished O’Handley, finding there was more than an “arms-length relationship” between the federal defendants and the social-media companies:

Plaintiffs allege a formal government-created system for federal officials to influence social-media censorship decisions. For example, the Complaint alleges that federal officials set up a long series of formal meetings to discuss censorship, setting up privileged reporting channels to demand censorship, and funding and establishing federal-private partnership to procure censorship of disfavored viewpoints. The Complaint clearly alleges that Defendants specifically authorized and approved the actions of the social-media companies and gives dozens of examples where Defendants dictated specific censorship decisions to social-media platforms. These allegations are a far cry from the complained-of action in O’Handley: a single message from an unidentified member of a state agency to Twitter.[161]

Finally, the court also found similarities between Skinner and Missouri v Biden that would support a finding of state action:

Section 230 of the CDA purports to preempt state laws to the contrary, thus removing all legal barriers to the censorship immunized by Section 230. Federal officials have also made plain a strong preference and desire to “share the fruits of such intrusions,” showing “clear indices of the Government’s encouragement, endorsement, and participation” in censorship, which “suffice to implicate the [First] Amendment.”

The Complaint further explicitly alleges subsidization, authorization, and preemption through Section 230, stating: “[T]hrough Section 230 of the Communications Decency Act (CDA) and other actions, the federal government subsidized, fostered, encouraged, and empowered the creation of a small number of massive social-media companies with disproportionate ability to censor and suppress speech on the basis of speaker, content, and viewpoint.” Section 230 immunity constitutes the type of “tangible financial aid,” here worth billions of dollars per year, that the Supreme Court identified in Norwood, 413 U.S. at 466, 93 S.Ct. 2804. This immunity also “has a significant tendency to facilitate, reinforce, and support private” censorship. Id. Combined with other factors such as the coercive statements and significant entwinement of federal officials and censorship decisions on social-media platforms, as in Skinner, this serves as another basis for finding government action.[162]

Again, there is tension in the opinions of these cases on the intersection of social media and the First Amendment under the joint-action or symbiotic-relationship test. But there are ways to read the cases consistently. First, there were far more factual allegations in Missouri v. Biden relative to the O’Handley, Hart, or Children’s Health Defense cases, particularly regarding how involved the federal defendants were in prodding social-media companies to moderate misinformation. There is even a way to read the different legal conclusions on Section 230 and Skinner consistently. The court in Biden v. Missouri made clear that it wasn’t Section 230 alone that made it like Skinner, but the combination of Section 230 immunity with other factors present:

The Defendants’ alleged use of Section 230’s immunity—and its obvious financial incentives for social-media companies—as a metaphorical carrot-and-stick combined with the alleged back-room meetings, hands-on approach to online censorship, and other factors discussed above transforms Defendants’ actions into state action. As Defendants note, Section 230 was designed to “reflect a deliberate absence of government involvement in regulating online speech,” but has instead, according to Plaintiffs’ allegations, become a tool for coercion used to encourage significant joint action between federal agencies and social-media companies.[163]

While there could be dangers inherent in treating Section 230 alone as an argument that social-media companies are state actors, the court appears inclined to say it is not Section 230 but rather the threat of removing it, along with the other dealings and communications from the federal government, that makes this state action.

Under the law & economics theory outlined in Part I, the joint-action or symbiotic-relationship test is also an important exception to the general dichotomy between private and state action. In particular, it is important to deter state officials from engaging in surreptitious speech regulation by covertly interjecting themselves into social-media companies’ moderation decisions. The allegations in Missouri v. Biden, if proven true, do appear to outline a vast and largely hidden infrastructure through which federal officials use backchannels to routinely discuss social-media companies’ moderation decisions and often pressure them into removing disfavored content in the name of misinformation. This kind of government intervention into the “marketplace of ideas” and the “market for private speech governance” takes away companies’ ability to respond freely to market incentives in moderating misinformation, and replaces their own editorial discretion with the opinions of government officials.

III.    Applying the First Amendment to Government Regulation of Online Misinformation

A number of potential consequences might stem from a plausible claim of state action levied against online platforms using one of the theories described above. Part III.A will explore the likely result, which is that a true censorship-by-deputization scheme enacted through social-media companies would be found to violate the First Amendment. Part III.B will consider the question of remedies: even if there is a First Amendment violation, those whose content or accounts have been removed may not be restored. Part III.C will then offer alternative ways for the government to deal with the problem of online misinformation without offending the First Amendment.

A.      If State Action Is Found, Removal of Content Under Misinformation Policies Would Violate the First Amendment

At a high level, First Amendment jurisprudence does allow for government regulation of speech in limited circumstances. In those cases, the threshold question is whether the type of speech at issue is protected speech and whether the regulation is content-based.[164] If it is, then the government must show the state action is narrowly tailored to a compelling governmental interest: this is the so-called “strict scrutiny” standard.^[165] A compelling governmental interest is the highest interest the state has, something considered necessary or crucial, and beyond simply legitimate or important.^[166] “Narrow tailoring” means the regulation uses the least-restrictive means “among available, effective alternatives.”^[167] While not an impossible standard for the government to reach, “[s]trict scrutiny leave[s] few survivors.”^[168] Moreover, prior restraints of speech, which are defined as situations where speech is restricted before publication, are presumptively unconstitutional.[169]

Only for content- and viewpoint-neutral “time, place, and manner restrictions” will regulation of protected speech receive less than strict scrutiny.^[170] In those cases, as long as the regulation serves a “significant” government interest, and there are alternative channels available for the expression, the regulation is permissible.[171]

There are also situations where speech regulation—whether because the regulation aims at conduct but has speech elements or because the speech is not fully protected for some other reason—receives “intermediate scrutiny.”^[172] In those cases, the government must show the state action is narrowly tailored to an important or substantial governmental interest, and burdens no more speech than necessary.^[173] Beyond the levels of scrutiny to which speech regulation is subject, state actions involving speech also may be struck down for overbreadth[174] or vagueness.[175] Together, these doctrines work to protect a very large sphere of speech, beyond what is protected in most jurisdictions around the world.

The initial question that arises with alleged misinformation is how to even define it. Neither social-media companies nor the government actors on whose behalf they may be acting are necessarily experts in misinformation. This can result in “void-for-vagueness” problems.

In Høeg v. Newsom,[176] the U.S. District Court for the Eastern District of California considered California’s state law AB 2098, which would charge medical doctors with “unprofessional conduct” and subject them to discipline if they shared with patients “false information that is contradicted by contemporary scientific consensus contrary to the standard of care” as part of treatment or advice.[177] The court stated that “[a] statute is unconstitutionally vague when it either ‘fails to provide a person of ordinary intelligence fair notice of what is prohibited, or is so standardless that it authorizes or encourages seriously discriminatory enforcement’”[178] and that “[v]ague statutes are particularly objectionable when they ‘involve sensitive areas of First Amendment freedoms” because “they operate to inhibit the exercise of those freedoms.’”[179] The court rejected the invitation to apply a lower vagueness standard typically used for technical language because “contemporary scientific consensus” has no established technical meaning in the scientific community.[180] The court also asked a series of questions that would be particularly relevant to social-media companies acting on behalf of government actors in efforts to combat misinformation:

[W]ho determines whether a consensus exists to begin with? If a consensus does exist, among whom must the consensus exist (for example practicing physicians, or professional organizations, or medical researchers, or public health officials, or perhaps a combination)? In which geographic area must the consensus exist (California, or the United States, or the world)? What level of agreement constitutes a consensus (perhaps a plurality, or a majority, or a supermajority)? How recently in time must the consensus have been established to be considered “contemporary”? And what source or sources should physicians consult to determine what the consensus is at any given time (perhaps peer-reviewed scientific articles, or clinical guidelines from professional organizations, or public health recommendations)?[181]

The court noted that defining the consensus with reference to pronouncements from the U.S. Centers for Disease Control and Prevention or the World Health Organization would be unhelpful, as those entities changed their recommendations on several important health issues over the course of the COVID-19 pandemic:

Physician plaintiffs explain how, throughout the course of the COVID-19 pandemic, scientific understanding of the virus has rapidly and repeatedly changed. (Høeg Decl. ¶¶ 15-29; Duriseti Decl. ¶¶ 7-15; Kheriaty Decl. ¶¶ 7-10; Mazolewski Decl. ¶¶ 12-13.) Physician plaintiffs further explain that because of the novel nature of the virus and ongoing disagreement among the scientific community, no true “consensus” has or can exist at this stage. (See id.) Expert declarant Dr. Verma similarly explains that a “scientific consensus” concerning COVID-19 is an illusory concept, given how rapidly the scientific understanding and accepted conclusions about the virus have changed. Dr. Verma explains in detail how the so-called “consensus” has developed and shifted, often within mere months, throughout the COVID-19 pandemic. (Verma Decl. ¶¶ 13-42.) He also explains how certain conclusions once considered to be within the scientific consensus were later proved to be false. (Id. ¶¶ 8-10.) Because of this unique context, the concept of “scientific consensus” as applied to COVID-19 is inherently flawed.[182]

As a result, the court determined that “[b]ecause the term ‘scientific consensus’ is so ill-defined, physician plaintiffs are unable to determine if their intended conduct contradicts the scientific consensus, and accordingly ‘what is prohibited by the law.’”[183] The court upheld a preliminary injunction against the law because of a high likelihood of success on the merits.[184]

Assuming the government could define misinformation in a way that wasn’t vague, the next question is what level of First Amendment scrutiny would such edicts receive? It is clear for several reasons that regulation of online misinformation would receive, and fail, the highest form of constitutional scrutiny.

First, the threat of government censorship of speech through social-media misinformation policies could be considered a prior restraint. Prior restraints occur when the government (or actors on their behalf) restrict speech before publication. As the Supreme Court has put it many times, “any system of prior restraints of expression comes to this Court bearing a heavy presumption against its constitutional validity.”[185]

In Missouri v. Biden, the court found the plaintiffs had plausibly alleged prior restraints against their speech, and noted that “[t]hreatening penalties for future speech goes by the name of ‘prior restraint,’ and a prior restraint is the quintessential first-amendment violation.”[186] The court found it relevant that social-media companies could “silence” speakers’ voices at a “mere flick of the switch,”[187] and noted this could amount to “a prior restraint by preventing a user of the social-media platform from voicing their opinion at all.”[188] The court further stated that “bans, shadow-bans, and other forms of restrictions on Plaintiffs’ social-media accounts, are… de facto prior restraints, [a] clear violation of the First Amendment.”[189]

Second, it is clear that any restriction on speech based upon its truth or falsity would be a content-based regulation, and likely a viewpoint-based regulation, as it would require the state actor to take a side on a matter of dispute.[190] Content-based regulation requires strict scrutiny, and a reasonable case can be made that viewpoint-based regulation of speech is per se inconsistent with the First Amendment.[191]

In Missouri v. Biden, the court noted that “[g]overnment action, aimed at the suppression of particular views on a subject which discriminates on the basis of viewpoint, is presumptively unconstitutional.”[192] The court found that “[p]laintiffs allege a regime of censorship that targets specific viewpoints deemed mis-, dis-, or malinformation by federal officials. Because Plaintiffs allege that Defendants are targeting particular views taken by speakers on a specific subject, they have alleged a clear violation of the First Amendment, i.e., viewpoint discrimination.”[193]

Third, even assuming there is clearly false speech that government agents (and social-media companies acting on their behalf) could identify, false speech presumptively receives full First Amendment protection. In United States v. Alvarez^[194] the Supreme Court stated that while older cases may have stated that false speech does not receive full protection, those were “confined to the few ‘historic and traditional categories [of expression] long familiar to the bar.’”^[195] In other words, there was no “general exception to the First Amendment for false statements.”^[196] Thus, as protected speech, any regulation of false speech, as such, would run into strict scrutiny.

In order to survive First Amendment scrutiny, government agents acting through social-media companies would have to demonstrate a parallel or alternative justification to regulate the sort of low-value speech the Supreme Court has recognized as outside the protection of the First Amendment.[197] These exceptions include defamation, fraud, the tort of false light, false statements to government officials, perjury, falsely representing oneself as speaking for the government (and impersonation), and other similar examples of fraud or false speech integral to criminal conduct.^[198]

But the Alvarez Court noted that, even in areas where false speech does not receive protection, such as fraud and defamation, the Supreme Court has found the First Amendment requires that claims of fraud be based on more than falsity alone.^[199]

When it comes to fraud,[200] for instance, the Supreme Court has repeatedly noted that the First Amendment offers no protection.[201] But “[s]imply labeling an action one for ‘fraud’… will not carry the day.”[202] Prophylactic rules aimed at protecting the public from the (sometimes fraudulent) solicitation of charitable donations, for instance, have been found to be unconstitutional prior restraints on several occasions by the Court.[203] The Court has found that “in a properly tailored fraud action the State bears the full burden of proof. False statement alone does not subject a fundraiser to fraud liability… Exacting proof requirements… have been held to provide sufficient breathing room for protected speech.”[204]

As for defamation,^[205] the Supreme Court found in New York Times v. Sullivan^[206] that “[a]uthoritative interpretations of the First Amendment guarantees have consistently refused to recognize an exception for any test of truth—whether administered by judges, juries, or administrative officials—and especially one that puts the burden of proving truth on the speaker.”^[207] In Sullivan, the Court struck down an Alabama defamation statute, finding that in situations dealing with public officials, the mens rea must be actual malice: knowledge that the statement was false or reckless disregard for whether it was false.^[208]

Since none of these exceptions would apply to online misinformation dealing with medicine or election law, social-media companies’ actions on behalf of the government against such misinformation would likely fail strict scrutiny. While it is possible that a court would find protecting public health or election security to be a compelling interest, the government would still face great difficulty showing that a ban on false information is narrowly tailored. It is highly unlikely that a ban on false information, as such, will ever be the least-restrictive means of controlling a harm. As the Court put it in Alvarez:

The remedy for speech that is false is speech that is true… Freedom of speech and thought flows not from the beneficence of the state but from the inalienable rights of the person. And suppression of speech by the government can make exposure of falsity more difficult, not less so. Society has the right and civic duty to engage in open, dynamic, rational discourse. These ends are not well served when the government seeks to orchestrate public discussion through content-based mandates.[209]

As argued above in Part I, a vibrant marketplace of ideas requires that individuals have the ability to express their ideas, so that the best ideas win. This means counter-speech is better than censorship from government actors to help society determine what is true. The First Amendment’s protection against government intervention into the marketplace of ideas promotes a better answer to online misinformation. Thus, a finding that government actors can’t use social-media actors to censor, based on vague definitions of misinformation, through prior restraints and viewpoint discrimination, and aimed at protected speech, is consistent with an understanding of the world where information is dispersed.

B.      The Problem of Remedies for Social-Media ‘Censorship’: The First Amendment Still Only Applies to Government Action

There is a problem, however, for plaintiffs who win cases against social-media companies that are found to be state actors when they remove posts and accounts due to alleged misinformation: the remedies are limited.

First, once the state action is removed through injunction, social-media companies would be free to continue to moderate misinformation as they see fit, free from any plausible First Amendment claim. For instance, in Carlisle Communications, the 9th Circuit found that, once the state action was enjoined, the telecommunications company was again free to determine whether or not to extend its service to the plaintiff. As the court put it:

Mountain Bell insists that its new policy reflected its independent business judgment. Carlin argues that Mountain Bell was continuing to yield to state threats of prosecution. However, the factual question of Mountain Bell’s true motivations is immaterial.

This is true because, inasmuch as the state under the facts before us may not coerce or otherwise induce Mountain Bell to deprive Carlin of its communication channel, Mountain Bell is now free to once again extend its 976 service to Carlin. Our decision substantially immunizes Mountain Bell from state pressure to do otherwise. Should Mountain Bell not wish to extend its 976 service to Carlin, it is also free to do that. Our decision modifies its public utility status to permit this action. Mountain Bell and Carlin may contract, or not contract, as they wish.[210]

This is consistent with the district court’s actions in Missouri v. Biden. There, the court granted the motion for a preliminary injunction, but it only applied against government action and not against the social-media companies at all.[211] For instance, the injunction prohibits a number of named federal officials and agencies from:

(1) meeting with social-media companies for the purpose of urging, encouraging, pressuring, or inducing in any manner the removal, deletion, suppression, or reduction of content containing protected free speech posted on social-media platforms;

(2) specifically flagging content or posts on social-media platforms and/or forwarding such to social-media companies urging, encouraging, pressuring, or inducing in any manner for removal, deletion, suppression, or reduction of content containing protected free speech;

(3) urging, encouraging, pressuring, or inducing in any manner social-media companies to change their guidelines for removing, deleting, suppressing, or reducing content containing protected free speech;

(4) emailing, calling, sending letters, texting, or engaging in any communication of any kind with social-media companies urging, encouraging, pressuring, or inducing in any manner for removal, deletion, suppression ,or reduction of content containing protected free speech;

(5) collaborating, coordinating, partnering, switchboarding, and/or jointly working with the Election Integrity Partnership, the Virality Project, the Stanford Internet Observatory, or any like project or group for the purpose of urging, encouraging, pressuring, or inducing in any manner removal, deletion, suppression, or reduction of content posted with social-media companies containing protected free speech;

(6) threatening, pressuring, or coercing social-media companies in any manner to remove, delete, suppress, or reduce posted content of postings containing protected free speech;

(7) taking any action such as urging, encouraging, pressuring, or inducing in any manner social-media companies to remove, delete, suppress, or reduce posted content protected by the Free Speech Clause of the First Amendment to the United States Constitution;

(8) following up with social-media companies to determine whether the social-media companies removed, deleted, suppressed, or reduced previous social-media postings containing protected free speech;

(9) requesting content reports from social-media companies detailing actions taken to remove, delete, suppress, or reduce content containing protected free speech; and

(10) notifying social-media companies to Be on The Lookout (BOLO) for postings containing protected free speech.[212]

In other words, a social-media company would not necessarily even be required to reinstate accounts or posts of those who have been excluded under their misinformation policies. It would become a question of whether, responding to marketplace incentives sans government involvement, the social-media companies continue to find it in their interest to enforce such policies against those affected persons and associated content.

Another avenue for private plaintiffs may be with a civil rights claim under Section 1983.[213] If it can be proved that social-media companies participated in a joint action with government officials to restrict First Amendment rights, it may be possible to collect damages from them, as well as from government officials.[214] Plaintiffs may struggle, however, to prove compensatory damages, which would require proof of harm. Categories of harm like physical injury aren’t relevant to social-media moderation policies, leaving things like diminished earnings or impairment of reputation. In most cases, it is likely that the damages to plaintiffs are de minimis and hardly worth the expense of filing suit. To receive punitive damages, plaintiffs would have to prove “the defendant’s conduct is… motivated by evil motive or intent, or when it involves reckless or callous indifference to the federally protected rights of others.”[215] This seems like it would be difficult to establish against the social-media companies unless there was an admission in the record that those companies’ goal was to suppress rights, rather than that they were attempting in good faith to restrict misinformation or simply acceding to government inducements.

The remedies available for constitutional violations in claims aimed at government officials are consistent with a theory of the First Amendment that prioritizes protecting the marketplace of ideas from intervention. While it leaves many plaintiffs with limited remedies against the social-media companies once the government actions are enjoined or deterred, it does return the situation to one where the social-media companies can freely compete in a market for speech governance on misinformation, as well.

C.      What Can the Government Do Under the First Amendment in Response to Misinformation on Social-Media Platforms?

If direct government regulation or implicit intervention through coercion or collusion with social-media companies is impermissible, the question may then arise as to what, exactly, the government can do to combat online misinformation.

The first option was already discussed in Part III.A in relation to Alvarez and narrow tailoring: counter-speech. Government agencies concerned about health or election misinformation could use social=media platforms to get their own message out. Those agencies could even amplify and target such counter-speech through advertising campaigns tailored to those most likely to share or receive misinformation.

Similarly, government agencies could create their own apps or social-media platforms to publicize information that counters alleged misinformation. While this may at first appear to be an unusual step, the federal government does, through the Corporation for Public Broadcasting, subsidize public television and public radio. If there is a fear of online misinformation, creating a platform where the government can promote its own point of view could combat online misinformation in a way that doesn’t offend the First Amendment.

Additionally, as discussed above in Part II.B in relation to O’Handley and the distinction between convincing and coercion: the government may flag alleged misinformation and even attempt to persuade social-media companies to act, so long as such communications involve no implicit or explicit threats of regulation or prosecution if nothing is done. The U.S. District Court for the Western District of Louisiana distinguished between constitutional government speech and unconstitutional coercion or encouragement in its memorandum accompanying its preliminary injunction in Missouri v. Biden:

Defendants also argue that a preliminary injunction would restrict the Defendants’ right to government speech and would transform government speech into government action whenever the Government comments on public policy matters. The Court finds, however, that a preliminary injunction here would not prohibit government speech… The Defendants argue that by making public statements, this is nothing but government speech. However, it was not the public statements that were the problem. It was the alleged use of government agencies and employees to coerce and/or significantly encourage social-media platforms to suppress free speech on those platforms. Plaintiffs point specifically to the various meetings, emails, follow-up contacts, and the threat of amending Section 230 of the Communication Decency Act. Plaintiffs have produced evidence that Defendants did not just use public statements to coerce and/or encourage social-media platforms to suppress free speech, but rather used meetings, emails, phone calls, follow-up meetings, and the power of the government to pressure social-media platforms to change their policies and to suppress free speech. Content was seemingly suppressed even if it did not violate social-media policies. It is the alleged coercion and/or significant encouragement that likely violates the Free Speech Clause, not government speech, and thus, the Court is not persuaded by Defendants’ arguments here.[216]

As the court highlights, there is a special danger in government communications that remain opaque to the public. Requests for action from social-media companies on misinformation should all be public information and not conducted behind closed doors or in covert communications. Such transparency would make it much easier for the public and the courts to determine whether state actors are engaged in government speech or crossing the line into coercion or substantial encouragement to suppress speech.

On the other hand, laws like the recent SB 262 in Florida[217] go beyond the delicate First Amendment balance that courts have tried to achieve. That law would limit government officials’ ability to share any information with social-media companies regarding misinformation, limiting contacts to the removal of criminal content or accounts, or an investigation or inquiry to prevent imminent bodily harm, loss of life, or property damage.[218] While going beyond the First Amendment standard may be constitutional, these restrictions could be especially harmful when the government has information that may not be otherwise available to the public. As important as it is to restrict government intervention, it would harm the marketplace of ideas to prevent government participation altogether.

Finally, Section 230 reform efforts aimed at limiting immunity in instances where social-media companies have “red flag” knowledge of defamatory material would be another constitutional way to address misinformation.[219] For instance, if a social-media company was presented with evidence that a court or arbitrator finds certain statements to be untrue, it could be required to make reasonable efforts to take down such misinformation, and keep it down.

Such a proposal would have real-world benefits. For instance, in the recent litigation brought by Dominion Voting Systems against Fox News, the court found the various factual claims about Dominion rigging the election for Joseph Biden were false.[220] While there was no final finding of liability due to Fox and Dominion coming to a settlement,[221] if Dominion were to present the court’s findings to a social-media company, the company would, under this proposal, have an obligation to remove content that repeats the claims the court found to be false. Similarly, an arbitrator finding that MyPillow CEO Mike Lindell’s claims that he had evidence of Chinese interference in the election were demonstrably false[222] could be enough to have those claims removed, as well. Rudy Giuliani’s recent finding of liability for defamation against two Georgia election workers could similarly be removed.[223]

However, these benefits may be limited by the fact that not every defamation claim resolves with a court finding falsity of a statement. Some cases settle before it gets that far, and the underlying claims remain unproven allegations. And, as discussed above, defamation itself is not easy to prove, especially for public figures who must also be able to show “actual malice.”[224] As a result, many cases won’t even be brought. This means there could be quite a bit defamatory information put out into the world that courts or arbitrators are unlikely to have occasion to consider.

On the other hand, to make a social-media company responsible for removing allegedly defamatory information in the absence of some competent legal authority finding the underlying claim false could be ripe for abuses that could have drastic chilling effects on speech. Thus, any Section 230 reform must be limited to those occasions where a court or arbitrator of competent authority (and with some finality of judgment) has spoken on the falsity of a statement.

Conclusion

There is an important distinction in First Amendment jurisprudence between private and state action. To promote a free market in ideas, we must also protect private speech governance, like that of social-media companies. Private actors are best placed to balance the desires of people for speech platforms and the regulation of misinformation.

But when the government puts its thumb on the scale by pressuring those companies to remove content or users in the name of misinformation, there is no longer a free marketplace of ideas. The First Amendment has exceptions in its state-action doctrine that would allow courts to enjoin government actors from initiating coercion of or collusion with private actors to do that which would be illegal for the government to do itself. Government censorship by deputization is no more allowed than direct regulation of alleged misinformation.

There are, however, things the government can do to combat misinformation, including counter-speech and nonthreatening communications with social-media platforms. Section 230 could also be modified to require the takedown of adjudicated misinformation in certain cases.

At the end of the day, the government’s role in defining or policing misinformation is necessarily limited in our constitutional system. The production of true knowledge in the marketplace of ideas may not be perfect, but it is the least bad system we have yet created.

[1] West Virginia Bd. of Ed. v. Barnette, 319 U.S. 624, 642 (1943).

[2] United States v. Alvarez, 567 U.S. 709, 728 (2012).

[3] See Amanda Seitz, Disinformation Board to Tackle Russia, Migrant Smugglers, Associated Press (Apr. 28, 2022), https://apnews.com/article/russia-ukraine-immigration-media-europe-misinformation-4e873389889bb1d9e2ad8659d9975e9d.

[4] See, e.g., Rep. Doug Lamafa, Brave New World? Orwellian ‘Disinformation Governance Board’ Goes Against Nation’s Principles, The Hill (May 4, 2022), https://thehill.com/opinion/congress-blog/3476632-brave-new-world-orwellian-disinformation-governance-board-goes-against-nations-principles; Letter to Secretary Mayorkas from Ranking Members of the House Committee on Oversight and Reform (Apr. 29, 2022), available at https://oversight.house.gov/wp-content/uploads/2022/04/Letter-to-DHS-re-Disinformation-Governance-Board-04292022.pdf (stating “DHS is creating the Orwellian-named “Disinformation Governance Board”); Jon Jackson, Joe Biden’s Disinformation Board Likened to Orwell’s ‘Ministry of Truth’, Newsweek (Apr. 29, 2022), https://www.newsweek.com/joe-bidens-disinformation-board-likened-orwells-ministry-truth-1702190.

[5] See Geneva Sands, DHS Shuts Down Disinformation Board Months After Its Efforts Were Paused, CNN (Aug. 24, 2022), https://www.cnn.com/2022/08/24/politics/dhs-disinformation-board-shut-down/index.html.

[6] For an example of this type of hearing, see Preserving Free Speech and Reining in Big Tech Censorship, Hearing before the U.S. House Energy and Commerce Subcommittee on Communications and Technology (Mar. 28, 2023), https://www.congress.gov/event/118th-congress/house-event/115561.

[7] See Ken Klippenstein & Lee Fang, Truth Cops: Leaked Documents Outline DHS’s Plans to Police Disinformation, The Intercept (Oct. 31, 2022), https://theintercept.com/2022/10/31/social-media-disinformation-dhs.

[8] See Matt Taibbi, Capsule Summaries of all Twitter Files Threads to Date, With Links and a Glossary, Racket News (last updated Mar. 17, 2023), https://www.racket.news/p/capsule-summaries-of-all-twitter. For evidence that Facebook received similar pressure from and/or colluded with government officials, see Robby Soave, Inside the Facebook Files: Emails Reveal the CDC’s Role in Silencing COVID-19 Dissent, reason (Jan. 19, 2023), https://reason.com/2023/01/19/facebook-files-emails-cdc-covid-vaccines-censorship; Ryan Tracy, Facebook Bowed to White House Pressure, Removed Covid Posts, Wall St. J. (Jul. 28, 2023), https://www.wsj.com/articles/facebook-bowed-to-white-house-pressure-removed-covid-posts-2df436b7.

[9] See Missouri, et al. v. Biden, et al., No. 23-30445 (5th Cir. Sept. 8, 2023), slip op. at 2-14, available at https://www.ca5.uscourts.gov/opinions/pub/23/23-30445-CV0.pdf. Hearing on the Weaponization of the Federal Government, Hearing Before the Select Subcomm. on the Weaponization of the Fed. Gov’t (Mar. 30, 2023) (written testimony of D. John Sauer), available at https://judiciary.house.gov/sites/evo-subsites/republicans-judiciary.house.gov/files/2023-03/Sauer-Testimony.pdf.

[10] See infra Part I.

[11] Manhattan Community Access Corp. v. Halleck, 139 S. Ct. 1921, 1928 (2019).

[12] Cf. Whitney v. California, 274 U.S. 357, 377 (1927) (Brandeis, J., concurring) (“If there be time to expose through discussion the falsehood and fallacies, to avert the evil by the processes of education, the remedy to be applied is more speech, not enforced silence”).

[13] See, e.g., Abrams v. United States, 250 U.S. 616, 630 (1919) (Holmes, J., dissenting) (“Persecution for the expression of opinions seems to me perfectly logical. If you have no doubt of your premises or your power and want a certain result with all your heart you naturally express your wishes in law and sweep away all opposition. To allow opposition by speech seems to indicate that you think the speech impotent, as when a man says that he has squared the circle, or that you do not care whole-heartedly for the result, or that you doubt either your power or your premises. But when men have realized that time has upset many fighting faiths, they may come to believe even more than they believe the very foundations of their own conduct that the ultimate good desired is better reached by free trade in ideas — that the best test of truth is the power of the thought to get itself accepted in the competition of the market, and that truth is the only ground upon which their wishes safely can be carried out. That at any rate is the theory of our Constitution. It is an experiment, as all life is an experiment. Every year if not every day we have to wager our salvation upon some prophecy based upon imperfect knowledge. While that experiment is part of our system I think that we should be eternally vigilant against attempts to check the expression of opinions that we loathe and believe to be fraught with death, unless they so imminently threaten immediate interference with the lawful and pressing purposes of the law that an immediate check is required to save the country.”).

[14] Whitney v. California, 274 U.S. 357, 377 (1927). See also, Alvarez, 567 U.S. at 727-28 (“The remedy for speech that is false is speech that is true. This is the ordinary course in a free society. The response to the unreasoned is the rational; to the uninformed, the enlightened; to the straight-out lie, the simple truth. The theory of our Constitution is ‘that the best test of truth is the power of the thought to get itself accepted in the competition of the market.’ The First Amendment itself ensures the right to respond to speech we do not like, and for good reason. Freedom of speech and thought flows not from the beneficence of the state but from the inalienable rights of the person. And suppression of speech by the government can make exposure of falsity more difficult, not less so. Society has the right and civic duty to engage in open, dynamic, rational discourse. These ends are not well served when the government seeks to orchestrate public discussion through content-based mandates.”) (citations omitted).

[15] See, e.g., Jonathan Peters, The “Sovereigns of Cyberspace” and State Action: The First Amendment’s Applications—or Lack Thereof—to Third-Party Platforms, 32 Berk. Tech. L. J. 989 (2017) .

[16] See id. at 990, 992 (2017) (emphasizing the need to “talk about the [state action doctrine] until we settle on a view both conceptually and functionally right.”) (citing Charles L. Black, Jr., The Supreme Court, 1966 Term—Foreword: “State Action,” Equal Protection, and California’s Proposition 14, 81 Harv. L. Rev. 69, 70 (1967)).

[17] Or, in the framing of some: to allow too much harmful speech, including misinformation, if it drives attention to the platforms for more ads to be served. See Karen Hao, How Facebook and Google Fund Global Misinformation, MIT Tech. Rev. (Nov. 20, 2021), https://www.technologyreview.com/2021/11/20/1039076/facebook-google-disinformation-clickbait.

[18] See, e.g., Thomas Sowell, Knowledge and Decisions (1980).

[19] That is to say, the marketplace will not perfectly remove misinformation, but will navigate the tradeoffs inherent in limiting misinformation without empowering any one individual or central authority to determine what is true.

[20] See, e.g., Halleck, 139 S. Ct. at 1928; Denver Area Ed. Telecommunications Consortium, Inc. v. FCC, 518 U.S. 727, 737 (1996) (plurality opinion); Hurley v. Irish-American Gay, Lesbian and Bisexual Group of Boston, Inc., 515 U.S. 557, 566 (1995); Hudgens v. NLRB, 424 U.S. 507, 513 (1976).

[21] See Part II below.

[22] For instance, a person could order a visitor to leave their home for saying something offensive and the police would, if called upon, help to eject them as trespassers. In general, courts will enforce private speech restrictions that governments could never constitutionally enact. See Mark D. Rosen, Was Shelley v. Kraemer Incorrectly Decided? Some New Answers, 95 Cal. L. Rev. 451, 458-61 (2007) (listing a number of cases where the holding of Shelley v. Kraemer that court enforcement of private agreements was state action did not extend to the First Amendment, meaning that private agreements to limit speech are enforced).

[23] Halleck, 139 S. Ct. at 1928, 1934 (citations omitted) (emphasis added).

[24] Id. at 1930.

[25] Id. at 1930-31.

[26] It is worth noting that application of the right to editorial discretion to social-media companies is a question that will soon be before the Supreme Court in response to common-carriage laws passed in Florida and Texas that would require carriage of certain speech. The 5th and 11th U.S. Circuit Courts of Appeal have come to opposite conclusions on this point. Compare NetChoice, LLC v. Moody, 34 F.4th 1196 (11th Cir. 2022) (finding the right to editorial discretion was violated by Florida’s common-carriage law) and NetChoice, LLC v. Paxton, 49 F.4th 439 (5th Cir. 2022) (finding the right to editorial discretion was not violated by Texas’ common-carriage law).

[27] Miami Herald Publishing Co. v. Tornillo, 418 U.S. 241, 256 (1974).

[28] See id. at 247-54.

[29] Id. at 255 (citing Columbia Broadcasting System, Inc. v. Democratic National Committee, 412 U. S. 94, 117 (1973)),

[30] 47 U.S.C. §230(c).

[31] For a further discussion, see generally Geoffrey A. Manne, Ben Sperry, & Kristian Stout, Who Moderates the Moderators?: A Law & Economics Approach to Holding Online Platforms Accountable Without Destroying the Internet, 49 Rutgers Computer & Tech. L. J. 26 (2022).

[32] Much of this section is adapted from Ben Sperry, An L&E Defense of the First Amendment’s Protection of Private Ordering, Truth on the Market (Apr. 23, 2021), https://truthonthemarket.com/2021/04/23/an-le-defense-of-the-first-amendments-protection-of-private-ordering.

[33] See F.A. Hayek, The Use of Knowledge in Society, 35 Am. Econ. Rev. 519 (1945).

[34] Id. at 520.

[35] See supra notes 13-14 and associated text. See also David Schultz, Marketplace of Ideas, First Amendment Encyclopedia, https://www.mtsu.edu/first-amendment/article/999/marketplace-of-ideas (last updated by Jun. 2017 by David L. Hudson) (noting the history of the “marketplace of ideas” justification by the Supreme Court for the First Amendment’s protection of free speech from government intervention); J.S. Mill, On Liberty, Ch. 2 (1859); John Milton, Areopagitica (1644).

[36] Without delving too far into epistemology, some argue that this is even the case in the scientific realm. See, e.g., Thomas Kuhn, The Structure of Scientific Revolutions (1962). Even according to the perspective that some things are universally true across time and space, they still amount to a tiny fraction of what we call human knowledge. “Information” may be a better term for what economists are actually talking about.

[37] The Supreme Court has recently affirmed that the government may not compel speech by businesses subject to public-accommodation laws. See 303 Creative LLC v. Elenis, No. 21-476, slip op. (Jun. 30, 2023), available at https://www.supremecourt.gov/opinions/22pdf/21-476_c185.pdf. The Court will soon also have to determine whether common-carriage laws can be applied to social-media companies consistent with the First Amendment in the NetChoice cases noted above. See supra note 26.

[38] Sowell, supra note 18, at 240.

[39] Even those whom we most trust to have considered opinions and an understanding of the facts may themselves experience “expert failure”—a type of market failure—that is made likelier still when government rules serve to insulate such experts from market competition. See generally Roger Koppl, Expert Failure (2018).

[40] See, e.g., West Virginia Bd. of Ed. v. Barnette, 319 U.S. 624, 642 (1943) (“If there is any fixed star in our constitutional constellation, it is that no official, high or petty, can prescribe what shall be orthodox in politics, nationalism, religion, or other matters of opinion or force citizens to confess by word or act their faith therein. If there are any circumstances which permit an exception, they do not now occur to us.”).

[41] See, e.g., Alvarez, 567 U.S. at 728 (“Permitting the government to decree this speech to be a criminal offense, whether shouted from the rooftops or made in a barely audible whisper, would endorse government authority to compile a list of subjects about which false statements are punishable. That governmental power has no clear limiting principle. Our constitutional tradition stands against the idea that we need Oceania’s Ministry of Truth.”).

[42] Cf. Halleck, 131 S. Ct. at 1930-31.

[43] For a good explanation, see Jamie Whyte, Polluting Words: Is There a Coasean Case to Regulate Offensive Speech?, ICLE White Paper (Sep. 2021), available at https://laweconcenter.org/wp-content/uploads/2021/09/Whyte-Polluting-Words-2021.pdf.

[44] R.H. Coase, The Problem of Social Cost, 3 J. L. & Econ. 1, 2 (1960) (“The traditional approach has tended to obscure the nature of the choice that has to be made. The question is commonly thought of as one in which A inflicts harm on B and what has to be decided is: how should we restrain A? But this is wrong. We are dealing with a problem of a reciprocal nature. To avoid the harm to B would inflict harm on A. The real question that has to be decided is: should A be allowed to harm B or should B be allowed to harm A? The problem is to avoid the more serious harm.”).

[45] See id. at 8-10.

[46] See generally David S. Evans & Richard Shmalensee, Matchmakers: The New Economics of Multisided Platforms (2016).

[47] For more on how and why social-media companies govern online speech, see Kate Klonick, The New Governors: The People, Rules, and Processes Governing Online Speech, 131 HARV. L. REV. 1598 (2018).

[48] See Kate Conger, Tiffany Hsu, & Ryan Mac, Elon Musk’s Twitter Faces Exodus of Advertisers and Executives, The New York Times (Nov. 1, 2022), https://www.nytimes.com/2022/11/01/technology/elon-musk-twitter-advertisers.html (“[A]dvertisers — which provide about 90 percent of Twitter’s revenue — are increasingly grappling with Mr. Musk’s ownership of the platform. The billionaire, who is meeting advertising executives in New York this week, has spooked some advertisers because he has said he would loosen Twitter’s content rules, which could lead to a surge in misinformation and other toxic content.”); Ryan Mac & Tiffany Hsu, Twitter’s US Ad Sales Plunge 59% as Woes Continue, The New York Times (Jun. 5, 2013), https://www.nytimes.com/2023/06/05/technology/twitter-ad-sales-musk.html (“Six ad agency executives who have worked with Twitter said their clients continued to limit spending on the platform. They cited confusion over Mr. Musk’s changes to the service, inconsistent support from Twitter and concerns about the persistent presence of misleading and toxic content on the platform.”).

[49] See, e.g., Brian Fung, Twitter Prepares to Roll Out New Paid Subscription Service That Includes Blue Checkmark, CNN (Nov. 5, 2022), https://www.cnn.com/2022/11/05/business/twitter-blue-checkmark-paid-subscription/index.html.

[50] Sowell, supra note 18, at 244.

[51] See Halleck, 139 S. Ct. at 1931 (“The Constitution does not disable private property owners and private lessees from exercising editorial discretion over speech and speakers on their property.”).

[52] Cf. Tornillo, 418 U.S. at 255 (“The power of a privately owned newspaper to advance its own political, social, and economic views is bounded by only two factors: first, the acceptance of a sufficient number of readers—and hence advertisers —to assure financial success; and, second, the journalistic integrity of its editors and publishers.”).

[53] See Ben Sperry & R.J. Lehmann, Gov. Desantis’ Unconstitutional Attack on Social Media, Tampa Bay Times (Mar. 3, 2021), https://www.tampabay.com/opinion/2021/03/03/gov-desantis-unconstitutional-attack-on-social-media-column (“Social-media companies and other tech platforms find themselves in a very similar position [as the newspaper in Tornillo] today. Just as newspapers do, Facebook, Google and Twitter have the right to determine what kind of content they want on their platforms. This means they can choose whether and how to moderate users’ news feeds, search results and timelines consistent with their own views on, for example, what they consider to be hate speech or misinformation. There is no obligation for them to carry speech they don’t wish to carry, which is why DeSantis’ proposal is certain to be struck down.”).

[54] See 47 U.S.C. §230.

[55] See, e.g., Jennifer Huddleston, Competition and Content Moderation: How Section 230 Enables Increased Tech Marketplace Entry, at 4, Cato Policy Analysis No. 922 (Jan. 31, 2022), available at https://www.cato.org/sites/cato.org/files/2022-01/policy-analysis-922.pdf (“The freedom to adopt content moderation policies tailored to their specific business model, their advertisers, and their target customer base allows new platforms to please internet users who are not being served by traditional media. In some cases, the audience that a new platform seeks to serve is fairly narrowly tailored. This flexibility to tailor content moderation policies to the specific platform’s community of users, which Section 230 provides, has made it possible for websites to establish online communities for a highly diverse range of people and interests, ranging from victims of sexual assault, political conservatives, the LGBTQ+ community, and women of color to religious communities, passionate stamp collectors, researchers of orphan diseases, and a thousand other affinity groups. Changing Section 230 to require websites to accept all comers, or to limit the ability to moderate content in a way that serves specific needs, would seriously curtail platforms’ ability to serve users who might otherwise be ignored by incumbent services or traditional editors.”). 

[56] See, e.g., Rui Gu, Lih-Bin Oh, & Kanliang Wang, Multi-Homing On SNSS: The Role of Optimum Stimulation Level and Perceived Complementarity in Need Gratification, 53 Information & Management 752 (2016), available at https://kd.nsfc.gov.cn/paperDownload/ZD19894097.pdf (“Given the increasingly intense competition for social networking sites (SNSs), ensuring sustainable growth in user base has emerged as a critical issue for SNS operators. Contrary to the common belief that SNS users are committed to using one SNS, anecdotal evidence suggests that most users use multiple SNSs simultaneously. This study attempts to understand this phenomenon of users’ multi-homing on SNSs. Building upon optimum stimulation level (OSL) theory, uses and gratifications theory, and literature on choice complementarity, a theoretical model for investigating SNS users’ multi-homing intention is proposed. An analysis of survey data collected from 383 SNS users shows that OSL positively affects users’ perceived complementarity between different SNSs in gratifying their four facets of needs, namely, interpersonal communication, self-presentation, information, and entertainment. Among the four dimensions of perceived complementarity, only interpersonal communication and information aspects significantly affect users’ intention to multi-home on SNSs. The results from this study offer theoretical and practical implications for understanding and managing users’ multi-homing use of SNSs.”).

[57] See, e.g., How Has Social Media Emerged as a Powerful Communication Medium, University Canada West Blog (Sep. 25, 2022), https://www.ucanwest.ca/blog/media-communication/how-has-social-media-emerged-as-a-powerful-communication-medium:

Social media has taken over the business sphere, the advertising sphere and additionally, the education sector. It has had a long-lasting impact on the way people communicate and has now become an integral part of their lives. For instance, WhatsApp has redefined the culture of IMs (instant messaging) and taken it to a whole new level. Today, you can text anyone across the globe as long as you have an internet connection. This transformation has not only been brought about by WhatsApp but also Facebook, Twitter, LinkedIn and Instagram. The importance of social media in communication is a constant topic of discussion.

Online communication has brought information to people and audiences that previously could not be reached. It has increased awareness among people about what is happening in other parts of the world. A perfect example of the social media’s reach can be seen in the way the story about the Amazon Rainforest fire spread. It started with a single post and was soon present on everyone’s newsfeed across different social media platforms.

Movements, advertisements and products are all being broadcasted on social media platforms, thanks to the increase in the social media users. Today, businesses rely on social media to create brand awareness as well as to promote and sell their products. It allows organizations to reach customers, irrespective of geographical boundaries. The internet has facilitated a resource to humankind that has unfathomable reach and benefits.

[58] Governmental intervention here could be particularly destructive if it leads to the imposition of “expert” opinions from insulated government actors from the “intelligence community.” Koppl, in his study on expert failure, described the situation as “the entangled deep state,” stating in relevant part:

The entangled deep state is an only partially hidden informal network linking the intelligence community, military, political parties, large corporations including defense contractors, and others. While the interests of participants in the entangled deep state often conflict, members of the deep state share a common interest in maintaining the status quo of the political system independently of democratic processes. Therefore, denizens of the entangled deep state may sometimes have an incentive to act, potentially in secret, to tamp down resistant voices and to weaken forces challenging the political status quo… The entangled deep state produces the rule of experts. Experts must often choose for the people because the knowledge on the basis of which choices are made is secret, and the very choice being made may also be a secret involving, supposedly, “national security.”… The “intelligence community” has incentives that are not aligned with the general welfare or with democratic process. Koppl, supra note 39, at 228, 230-31.

[59] Halleck, 139 S. Ct. at 1928 (internal citations omitted).

[60] 326 U.S. 501 (1946).

[61] Id. at 506.

[62] Id. at 509 (“When we balance the Constitutional rights of owners of property against those of the people to enjoy freedom of press and religion, as we must here, we remain mindful of the fact that the latter occupy a preferred position.”).

[63] 391 U.S. 308 (1968).

[64] See id. at 316-19. In particular, see id. at 318 (“The shopping center here is clearly the functional equivalent of the business district of Chickasaw involved in Marsh.”).

[65] See id. at 325.

[66] 407 U.S. 551 (1972).

[67] Id. at 562.

[68] Id.

[69] See id. at 568 (“[T]he courts properly have shown a special solicitude for the guarantees of the First Amendment, this Court has never held that a trespasser or an uninvited guest may exercise general rights of free speech on property privately owned and used nondiscriminatorily for private purposes only.”).

[70] Id. at 569.

[71] See id. at 570.

[72] 424 U.S. 507 (1976).

[73] Id. at 513.

[74] See id. at 516 (“Under what circumstances can private property be treated as though it were public? The answer that Marsh gives is when that property has taken on all the attributes of a town, i. e., `residential buildings, streets, a system of sewers, a sewage disposal plant and a “business block” on which business places are situated.’ (Logan Valley, 391 U.S. at 332 (Black, J. dissenting) (quoting Marsh, 326 U.S. at 502)).

[75] See id. at 518 (“It matters not that some Members of the Court may continue to believe that the Logan Valley case was rightly decided. Our institutional duty is to follow until changed the law as it now is, not as some Members of the Court might wish it to be. And in the performance of that duty we make clear now, if it was not clear before, that the rationale of Logan Valley did not survive the Court’s decision in the Lloyd case.”).

[76] Id. at 521.

[77] Jackson v. Metropolitan Edison Co., 419 U.S. 345, 352 (1974).

[78] See, e.g., the discussion about Prager University v. Google below.

[79] Packingham v. North Carolina, 137 S. Ct. 1730, 1737 (2017).

[80] Id. (internal citation omitted).

[81] See, e.g., Brock v. Zuckerberg, 2021 WL 2650070, at *3 (S.D.N.Y. Jun. 25, 2021); Freedom Watch, Inc. v. Google Inc., 816 F. App’x 497, 499 (D.C. Cir. 2020); Zimmerman v. Facebook, Inc., 2020 WL 5877863 at *2 (N.D. Cal. Oct. 2, 2020); Ebeid v. Facebook, Inc., 2019 WL 2059662 at *6 (N.D. Cal. May 9, 2019); Green v. YouTube, LLC, 2019 WL 1428890, at *4 (D.N.H. Mar. 13, 2019); Nyabwa v. FaceBook, 2018 WL 585467, at *1 (S.D. Tex. Jan. 26, 2018); Shulman v. Facebook.com, 2017 WL 5129885, at *4 (D.N.J. Nov. 6, 2017).

[82] Halleck, 139 S. Ct. at 1929 (emphasis in original).

[83] Id. at 1930.

[84] Id.

[85] Id. at 1930-31.

[86] 951 F.3d 991 (9th Cir. 2020).

[87] See id. at 997-98. See also, Prager University v. Google, LLC, 2018 WL 1471939, at *6 (N.D. Cal. Mar. 26, 2018) (“Plaintiff primarily relies on the United States Supreme Court’s decision in Marsh v. Alabama to support its argument, but Marsh plainly did not go so far as to hold that any private property owner “who operates its property as a public forum for speech” automatically becomes a state actor who must comply with the First Amendment.”).

[88] See PragerU, 951 F.3d at 996-99 (citing Halleck 12 times, Hudgens 3 times, and Tanner 3 times).

[89] See supra n. 7-9 and associated text.

[90] Cf. Norwood v. Harrison, 413 U.S. 455, 465 (1973) (“It is axiomatic that a state may not induce, encourage or promote private persons to accomplish what it is constitutionally forbidden to accomplish.”).

[91] Blum v. Yaretsky, 457 U.S. 991, 1004 (1982).

[92] Id. at 1004-05.

[93] Id. (noting that “the factual setting of each case will be significant”).

[94] 372 U.S. 58 (1963).

[95] See id. at 66-67.

[96] See id. at 68.

[97] Id. at 67.

[98] Id. at 68-69.

[99] 827 F.2d 1291 (9th Cir. 1987).

[100] See id. at 1295.

[101] Id.

[102] See id. (“Simply by ‘command[ing] a particular result,’ the state had so involved itself that it could not claim the conduct had actually occurred as a result of private choice.”) (quoting Peterson v. City of Greenville, 373 U.S. 244, 248 (1963)).

[103] See Backpage.com, LLC v. Dar, 807 F.3d 229 (7th Cir. 2015).

[104] See id. at 231, 232.

[105] Id. at 230.

[106] Id. at 235.

[107] Id. at 231.

[108] 2023 WL 2443073 (9th Cir. Mar. 10, 2023).

[109] See id. at *2-3.

[110] See id. at *5-6.

[111] Id. at *6.

[112] Id.

[113] Id.

[114] 2022 WL 1427507 (N.D. Cal. May 5, 2022).

[115] See id. at *8.

[116] Id.

[117] Id. (emphasis in original).

[118] See, e.g., Trump v. Twitter, 602 F.Supp.3d 1213, 1218-26 (2022); Children’s Health Def. v. Facebook, 546 F.Supp.3d 909, 932-33 (2021).

[119] 2023 WL 2578260 (W.D. La. Mar. 20, 2023). See also Missouri, et al. v. Biden, et al., 2023 WL 4335270 (W.D. La. Jul. 4., 2023) (memorandum opinion granting the plaintiffs’ motion for preliminary injunction).

[120] 2023 WL 2578260 at *30-31.

[121] See id.

[122] See id. at *17-19.

[123] It is worth noting that all of these cases were decided at the motion-to-dismiss stage, during which all of the plaintiffs’ allegations are assumed to be true. The plaintiffs in Missouri v. Biden will have to prove their factual case of state action. Now that the Western District of Louisiana has ruled on the motion for preliminary injunction, it is likely that there will be an appeal before the case gets to the merits.

[124] The district court in Missouri v. Biden discussed this distinction further in the memorandum ruling on request for preliminary injunction:

The Defendants argue that by making public statements, this is nothing but government speech. However, it was not the public statements that were the problem. It was the alleged use of government agencies and employees to coerce and/or significantly encourage social-media platforms to suppress free speech on those platforms. Plaintiffs point specifically to the various meetings, emails, follow-up contacts, and the threat of amending Section 230 of the Communication Decency Act. Plaintiffs have produced evidence that Defendants did not just use public statements to coerce and/or encourage social-media platforms to suppress free speech, but rather used meetings, emails, phone calls, follow-up meetings, and the power of the government to pressure social-media platforms to change their policies and to suppress free speech. Content was seemingly suppressed even if it did not violate social-media policies. It is the alleged coercion and/or significant encouragement that likely violates the Free Speech Clause, not government speech, and thus, the Court is not persuaded by Defendants’ arguments here.

Missouri v. Biden, 2023 WL 4335270, at *56 (W.D. La. July 4, 2023).

[125] While the district court did talk in significantly greater detail about specific allegations as to each federal defendant’s actions in coercing or encouraging changes in moderation policies or enforcement actions, there is still a lack of specificity as to how it affected the plaintiffs. See id. at *45-53 (applying the coercion/encouragement standard to each federal defendant). As in its earlier decision at the motion-to-dismiss stage, the court’s opinion accompanying the preliminary injunction does deal with this issue to a much greater degree in its discussion of standing, and specifically of traceability. See id. at *61-62:

Here, Defendants heavily rely upon the premise that social-media companies would have censored Plaintiffs and/or modified their content moderation policies even without any alleged encouragement and coercion from Defendants or other Government officials. This argument is wholly unpersuasive. Unlike previous cases that left ample room to question whether public officials’ calls for censorship were fairly traceable to the Government; the instant case paints a full picture. A drastic increase in censorship, deboosting, shadow-banning, and account suspensions directly coincided with Defendants’ public calls for censorship and private demands for censorship. Specific instances of censorship substantially likely to be the direct result of Government involvement are too numerous to fully detail, but a birds-eye view shows a clear connection between Defendants’ actions and Plaintiffs injuries.

The Plaintiffs’ theory of but-for causation is easy to follow and demonstrates a high likelihood of success as to establishing Article III traceability. Government officials began publicly threatening social-media companies with adverse legislation as early as 2018. In the wake of COVID-19 and the 2020 election, the threats intensified and became more direct. Around this same time, Defendants began having extensive contact with social-media companies via emails, phone calls, and in-person meetings. This contact, paired with the public threats and tense relations between the Biden administration and social-media companies, seemingly resulted in an efficient report-and-censor relationship between Defendants and social-media companies. Against this backdrop, it is insincere to describe the likelihood of proving a causal connection between Defendants’ actions and Plaintiffs’ injuries as too attenuated or purely hypothetical.

The evidence presented thus goes far beyond mere generalizations or conjecture: Plaintiffs have demonstrated that they are likely to prevail and establish a causal and temporal link between Defendants’ actions and the social-media companies’ censorship decisions. Accordingly, this Court finds that there is a substantial likelihood that Plaintiffs would not have been the victims of viewpoint discrimination but for the coercion and significant encouragement of Defendants towards social-media companies to increase their online censorship efforts.

[126] See Lugar v. Edmonson Oil Co., 457 U.S. 922, 941-42 (1982).

[127] See Brentwood Acad. v. Tennessee Secondary Sch. Athletic Ass’n, 531 U.S. 288, 294 (2001).

[128] See id. at 296.

[129] For instance, in Mathis v. Pacific Gas & Elec. Co., 75 F.3d 498 (9th Cir. 1996), the 9th Circuit described the plaintiff’s “joint action” theory as one where a private person could only be liable if the particular actions challenged are “inextricably intertwined” with the actions of the government. See id. at 503.

[130] See Brentwood, 531 U.S. at 296 (noting that “examples may be the best teachers”).

[131] See Lugar, 457 U.S. at 925.

[132] See id.

[133] See id.

[134] Id. at 941 (internal citations omitted).

[135] Id.

[136] See id. at 942.

[137] 365 U.S. 715 (1961).

[138] See id. at 717-20.

[139] Id. at 724.

[140] See Rendell-Baker v. Kohn, 457 U.S. 830, 842-43 (1982).

[141] See Brentwood, 531 U.S. at 292-93.

[142] See id. at 296 (“[A] challenged activity may be state action… when it is ‘entwined with governmental policies,’ or when government is ‘entwined in [its] management or control.’”) (internal citations omitted).

[143] See id. at 298-301.

[144] Id. at 302.

[145] 489 U.S. 602 (1989).

[146] See id. at 606-12, 615.

[147] Id. at 615.

[148] Id.

[149] O’Handley, 2023 WL 2443073, at *7.

[150] Id.

[151] See id. at *7-8.

[152] 75 F.3d 498 (9th Cir. 1996).

[153] O’Handley, 2023 WL 2443073, at *8.

[154] Id.

[155] Hart, 2022 WL 1427507, at *6.

[156] Id. at *7.

[157] See, e.g., Fed. Agency of News LLC v. Facebook, Inc., 432 F. Supp. 3d 1107, 1124-27 (N.D. Cal. 2020); Children’s Health Def. v. Facebook Inc., 546 F. Supp. 3d 909, 927-31 (N.D. Cal. 2021); Berenson v. Twitter, 2022 WL1289049, at *3 (N.D. Cal. Apr. 29, 2022).

[158] 546 F. Supp. 3d 909 (N.D. Cal. 2021).

[159] Id. at 932 (citing Divino Grp. LLC v. Google LLC, 2021 WL 51715, at *6 (N.D. Cal. Jan. 6, 2021)).

[160] Missouri v. Biden, 2023 WL 2578260, at *33.

[161] Id.

[162] Id. at *33-34.

[163] Id. at *34.

[164] A government action is content based if it can’t be applied without considering its content. See, e.g., Reed v. Town of Gilbert, Ariz., 576 U.S. 155, 163 (2015) (“Government regulation of speech is content based if a law applies to particular speech because of the topic discussed or the idea or message expressed.”).

^[165] See, e.g., Citizens United v. Fed. Election Comm’n, 558 U.S. 310, 340 (2010) (“Laws that burden political speech are ‘subject to strict scrutiny,’ which requires the Government to prove that the restriction ‘furthers a compelling interest and is narrowly tailored to achieve that interest.’”) (internal citations omitted).

^[166] See Fulton v. City of Philadelphia, Pennsylvania, 141 S. Ct. 1868, 1881 (2021) (“A government policy can survive strict scrutiny only if it advances ‘interests of the highest order’…”).

^[167] Ashcroft v. ACLU, 542 U.S. 656, 666 (2004). In that case, the Court compared the Children’s Online Protection Act’s age-gating to protect children from online pornography to blocking and filtering software available in the marketplace, and found those alternatives to be less restrictive. The Court thus struck down the regulation. See id. at 666-70.

^[168] Alameda Books v. City of Los Angeles, 535 U.S. 425, 455 (2002).

[169] See, e.g., New York Times Co. v. United States, 403 U.S. 713, 714 (1971).

^[170] The classic example being an ordinance on noise that doesn’t require the government actor to consider the content or viewpoint of the speaker in order to enforce. See Ward v. Rock Against Racism, 491 U.S. 781 (1989).

[171] See id. at 791 (“Our cases make clear, however, that even in a public forum the government may impose reasonable restrictions on the time, place, or manner of protected speech, provided the restrictions ‘are justified without reference to the content of the regulated speech, that they are narrowly tailored to serve a significant governmental interest, and that they leave open ample alternative channels for communication of the information.’”) (internal citations omitted).

^[172] See Turner Broad. Sys., Inc. v. FCC, 512 U.S. 622, 662 (1994) (finding “the appropriate standard by which to evaluate the constitutionality of must-carry is the intermediate level of scrutiny applicable to content-neutral restrictions that impose an incidental burden on speech.”).

^[173] See id. (“[A] content-neutral regulation will be sustained if ‘it furthers an important or substantial governmental interest; if the governmental interest is unrelated to the suppression of free expression; and if the incidental restriction on alleged First Amendment freedoms is no greater than is essential to the furtherance of that interest.’”) (quoting United States v. O’Brien, 391 U.S. 367, 377 (1968)).

[174] See Broadrick v. Oklahoma, 413 U.S. 601, 615 (1973) (holding that “the overbreadth of a statute must not only be real, but substantial as well, judged in relation to the statute’s plainly legitimate sweep”).

[175] See Kolender v. Lawson, 461 U.S. 352, 357 (1983) (holding that a law must have “sufficient definiteness that ordinary people can understand what conduct is prohibited and in a manner that does not encourage arbitrary and discriminatory enforcement”).

[176] 2023 WL 414258 (E.D. Cal. Jan. 25, 2023).

[177] Cal. Bus. & Prof. Code § 2270.

[178] Høeg, 2023 WL 414258, at *6 (internal citations omitted).

[179] Id. at *7.

[180] See id.

[181] Id. at *8.

[182] Id. at *9.

[183] Id. at *9.

[184] See id. at *12.

[185] New York Times Co. v. United States, 403 U.S. 713, 714 (1971) (quoting Bantam Books, 372 U.S. at 70).

[186] Missouri v. Biden, 2023 WL2578260, at *35 (quoting Backpage.com, 807 F.3d at 230).

[187] See id. (comparing the situation to cable operators in the Turner Broadcasting cases).

[188] Id.

[189] Id.

[190] See discussion of United States v. Alvarez, 567 U.S. 709 (2012) below.

[191] See Minnesota Voters Alliance v. Mansky, 138 S. Ct. 1876, 1885 (2018) (“In a traditional public forum — parks, streets, sidewalks, and the like — the government may impose reasonable time, place, and manner restrictions on private speech, but restrictions based on content must satisfy strict scrutiny, and those based on viewpoint are prohibited.”).

[192] Missouri v. Biden, 2023 WL2578260, at *35.

[193] Id.

^[194] 567 U.S. 709 (2012).

^[195] Id. at 717 (quoting United States v. Stevens, 559 U.S. 460, 468 (2010)).

^[196] Id. at 718.

[197] See Chaplinsky v. New Hampshire, 315 U.S. 568, 571-72 (1942) (“There are certain well-defined and narrowly limited classes of speech, the prevention and punishment of which has never been thought to raise any Constitutional problem.”)

^[198] See Alvarez, 567 U.S. at 718-22.

^[199] See id. at 719 (“Even when considering some instances of defamation and fraud, moreover, the Court has been careful to instruct that falsity alone may not suffice to bring the speech outside the First Amendment. The statement must be a knowing or reckless falsehood.”). This means that the First Amendment was found to limit common law actions against false speech which did not receive constitutional protection.

[200] Under the common law, the elements of fraud include (1) a misrepresentation of a material fact or failure to disclose a material fact the defendant was obligated to disclose, (2) intended to induce the victim to rely on the misrepresentation or omission, (3) made with knowledge that the statement or omission was false or misleading, (4) the plaintiff relied upon the representation or omission, and (5) suffered damages or injury as a result of the reliance. See, e.g., Mandarin Trading Ltd v. Wildenstein, 919 N.Y.S.2d 465, 469 (2011); Kostryckyj v. Pentron Lab. Techs., LLC, 52 A.3d 333, 338-39 (Pa. Super. 2012); Masingill v. EMC Corp., 870 N.E.2d 81, 88 (Mass. 2007). Similarly, commercial speech regulation on deceptive or misleading advertising or health claims have also been found to be consistent with the First Amendment. See Virginia State Bd. of Pharmacy v. Virginia Citizens Consumer Council, 425 U.S. 748, 771-72 (1976) (“Obviously, much commercial speech is not provably false, or even wholly false, but only deceptive or misleading. We foresee no obstacle to a State’s dealing effectively with this problem. The First Amendment, as we construe it today does not prohibit the State form insuring that the stream of commercial information flow cleanly as well as freely.”).

[201] See, e.g., Donaldson v. Read Magazine, Inc. 333 U.S. 178, 190 (1948) (the government’s power “to protect people against fraud” has “always been recognized in this country and is firmly established”).

[202] Illinois, ex rel. Madigan v. Telemarketing Associates, Inc., 538 U.S. 600, 617 (2003).

[203] See, e.g., Schaumburg v. Citizens for a Better Environment, 444 U.S. 620 (1980); Secretary of State of Md. v. Joseph H. Munson Co., 467 U.S. 947 (1984); Riley v. National Federation of Blind of N. C., Inc., 487 U.S. 781 (1988).

[204] Madigan, 538 U.S. at 620.

^[205] Under the old common-law rule, proving defamation required a plaintiff to present a derogatory statement and demonstrate that it could hurt their reputation. The falsity of the statement was presumed, and the defendant had the burden to prove the statement was true in all of its particulars. Re-publishing something from someone else could also open the new publisher to liability. See generally Samantha Barbas, The Press and Libel Before New York Times v. Sullivan, 44 Colum. J.L. & Arts 511 (2021).

^[206] 376 U.S. 254 (1964).

^[207] Id. at 271. See also id. at 271-72 (“Erroneous statement is inevitable in free debate, and [] it must be protected if the freedoms of expression are to have the ‘breathing space that they need to survive.’”) (quoting N.A.A.C.P. v. Button, 371 U.S. 415, 433 (1963)).

^[208] Id. at 279-80.

[209] Id. at 727-28.

[210] Carlin Commc’ns, 827 F.2d at 1297.

[211] See Missouri, et al. v. Biden, et al., Case No. 3:22-CV-01213 (W.D. La. Jul. 4, 2023), available at https://int.nyt.com/data/documenttools/injunction-in-missouri-et-al-v/7ba314723d052bc4/full.pdf.

[212] Id. See also Missouri, et al. v. Biden, et al., 2023 WL 4335270, at *45-56 (W.D. La. Jul. 4., 2023) (memorandum ruling on request for preliminary injunction). But see Missouri, et al. v. Biden, et al., No. 23-30445 (5th Cir. Sept. 8, 2023), slip op., available at https://www.ca5.uscourts.gov/opinions/pub/23/23-30445-CV0.pdf (upholding the injunction but limiting the parties it applies to); Murthy et al. v. Missouri, et al., No: 3:22-cv-01213 (Sept. 14, 2023) (order issued by Justice Aliso issuing an administrative stay of the preliminary injunction until Sept. 22, 2023 at 11:509 p.m. EDT).

[213] 42 U.S.C. §1983.

[214] See, e.g., Adickes v. SH Kress & Co., 398 U.S. 144, 152 (1970) (“Although this is a lawsuit against a private party, not the State or one of its officials, our cases make clear that petitioner will have made out a violation of her Fourteenth Amendment rights and will be entitled to relief under § 1983 if she can prove that a Kress employee, in the course of employment, and a Hattiesburg policeman somehow reached an understanding to deny Miss Adickes service in the Kress store, or to cause her subsequent arrest because she was a white person in the company of Negroes. The involvement of a state official in such a conspiracy plainly provides the state action essential to show a direct violation of petitioner’s Fourteenth Amendment equal protection rights, whether or not the actions of the police were officially authorized, or lawful… Moreover, a private party involved in such a conspiracy, even though not an official of the State, can be liable under § 1983.”) (internal citations omitted).

[215] Smith v. Wade, 461 U.S. 30, 56 (1983).

[216] See Missouri, et al. v. Biden, et al., 2023 WL 4335270, at *55, 56 (W.D. La. Jul. 4., 2023).

[217] Codified at Fla. Stat. § 112.23, available at https://casetext.com/statute/florida-statutes/title-x-public-officers-employees-and-records/chapter-112-public-officers-and-employees-general-provisions/part-i-conditions-of-employment-retirement-travel-expenses/section-11223-government-directed-content-moderation-of-social-media-platforms-prohibited.

[218] Id.

[219] For more on this proposal, Manne, Stout, & Sperry, supra note 31, at 106-112.

[220] See Dominion Voting Sys. v. Fox News Network, LLC, C.A. No.: N21C-03-257 EMD (Sup. Ct. Del. Mar. 31, 2023), available at https://www.documentcloud.org/documents/23736885-dominion-v-fox-summary-judgment.

[221] See, e.g.,  Jeremy W. Peters & Katie Robertson, Fox Will Pay $787.5 Million to Settle Defamation Suit, New York Times (Apr. 18, 2023), https://www.nytimes.com/live/2023/04/18/business/fox-news-dominion-trial-settlement#fox-dominion-defamation-settle.

[222] See, e.g., Neil Vigdor, ‘Prove Mike Wrong’ for $5 Million, Lindell Pitched. Now, He’s Told to Pay Up., New York Times (Apr. 20, 2023), https://www.nytimes.com/2023/04/20/us/politics/mike-lindell-arbitration-case-5-million.html.

[223] See Stephen Fowler, Judge Finds Rudy Giuliani Liable for Defamation of Two Georgia Election Workers, national public radio (Aug. 30, 2023), https://www.npr.org/2023/08/30/1196875212/judge-finds-rudy-giuliani-liable-for-defamation-of-two-georgia-election-workers.

[224] See supra notes 206-09 and associated text.

It seems that large language models (LLMs) are all the rage right now, from Bing’s announcement that it plans to integrate the ChatGPT technology into its search engine to Google’s announcement of its own LLM called “Bard” to Meta’s recent introduction of its Large Language Model Meta AI, or “LLaMA.” Each of these LLMs use artificial intelligence (AI) to create text-based answers to questions.

Read the full piece here.

SUMMARY OF ARGUMENT

An interactive computer service’s automated recommendations qualify for statutory immunity under Section 230(c)(1). Congress enacted this policy choice in clear text, supported by powerful statutory context, including express findings and purposes that it wrote into the statute itself. And Congress did so in service of a national policy favoring free and open discourse on the still developing internet—a policy that has proved enormously successful in the years since. This Court should resist Petitioners’ invitation to impose sweeping changes on the Nation’s internet policy, and instead leave any such changes if they ever prove necessary—to Congress.

Section 230’s text should decide this case. Section 230(c)(1) immunizes the user or provider of an “interactive computer service” from being “treated as the publisher or speaker” of information “provided by another information content provider.” And, as Section 230(f)’s definitions make clear, Congress understood the term “interactive computer service” to include services that “filter,” “screen,” “pick, choose, analyze,” “display, search, subset, organize,” or “reorganize” third-party content. Automated recommendations perform exactly those  functions, and are therefore within the express scope of Section 230’s text.

Section 230(c)(1)’s use of the phrase “treated as the publisher or speaker” further confirms that Congress immunized distributors of third-party information from liability. At common law, a distributor of third-party information could be held liable only when the doctrine permitted the distributor to be treated as the publisher. As Petitioners and the United States agree, Congress understood and incorporated that common-law meaning of “treated as the publisher” into Section 230(c)(1). Given that a distributor cannot be “treated as the publisher” of certain third-party information, however, there is no alternative mechanism for holding the distributor liable based on the improper character of the information. Indeed, Congress enacted Section 230(c)(1) specifically to avoid the sweeping consequences that the common-law regime of knowledge-based distributor liability would inflict on the developing internet.

Section 230(c)(1)’s surrounding and subsequent statutory context bolsters this conclusion. Section 230(c)(1) provides the same protection to “user[s]” as to “provider[s]” of interactive computer services. Petitioners do not defend the position that users who like, retweet, or otherwise amplify third-party content should be held liable for the character of that content, but Section 230(c)(1)’s text renders that an inescapable consequence of their argument. The better inference is that Congress chose to protect a wide range of speech and speech-promoting conduct for providers and users of interactive computer services alike. In addition, other statutory enactments illustrate that Congress knew how to impose liability on distributors when it wanted to—such as in the Digital Millennium Copyright Act, for example, where Congress also wrote a detailed notice-and-takedown framework into the statute to ensure that distributors received adequate procedural protections as well.

Petitioners’ and the United States’ attempts to distinguish between mere automated recommendations (for which distributors purportedly could be liable) and the recommended content (for which they could not) find no support in the text. To the contrary, the text makes clear that even a bare automated recommendation constitutes “pick[ing]” or “choos[ing]” content, an activity expressly contemplated by Section 230. Moreover, to hold a distributor liable based in part upon the improper content of information created by a third party would conflict with the common-law meaning of the terms Congress chose.

Congress enacted Section 230(c)(1) to protect the continuing development of the internet and ensure that it would remain a national forum for the free exchange of ideas. This is a case where the statutory text successfully implements Congress’s purposes by providing broad protections to automated recommendations of third-party information. But this Court need not guess at Congress’s purposes here, as it might be reluctant to do in a different case, because Congress enacted its purposes into the statute itself. Those purposes are part of the statutory text like any other statutory text, and deserving of the respect this Court would give to any text that passed through bicameralism and presentment into law. If any changes to our Nation’s statutory regulation of the internet are necessary, this Court should leave them to Congress.

Read the full brief here.

Later next month, the U.S. Supreme Court will hear oral arguments in Gonzalez v. Google LLC, a case that has drawn significant attention and many bad takes regarding how Section 230 of the Communications Decency Act should be interpreted. Enacted in the mid-1990s, when the Internet as we know it was still in its infancy, Section 230 has grown into a law that offers online platforms a fairly comprehensive shield against liability for the content that third parties post to their services. But the law has also come increasingly under fire, from both the political left and the right.

Read the full piece here.

“Just when I thought I was out, they pull me back in!” says Al Pacino’s character, Michael Corleone, in Godfather III. That’s how Facebook and Google must feel about S. 673, the Journalism Competition and Preservation Act (JCPA).

Read the full piece here.

Background…

As leaders of the U.S. Senate work to pass the National Defense Authorization Act (NDAA) in the ongoing lame-duck session, some reports suggest that S. 673, the Journalism Competition and Preservation Act (JCPA), could be added to the legislative package. Approved in September 2022 by the Senate Judiciary Committee, the JCPA aims to boost the fortunes of traditional media companies by forcing “covered” online platforms to pay for digital journalism accessed via their services. The bill would require that platforms continue to display digital journalism, while setting out an intricate process whereby digital-journalism providers would collectively negotiate the price of content with platforms.

But…

This quixotic attempt to prop up flailing media firms will create legally sanctioned cartels that harm consumers, while forcing online platforms to carry and pay for content in ways that violate long-established principles of intellectual property, economic efficiency, and the U.S. Constitution.

Read the full explainer here.

Executive Summary

The Federal Trade Commission (“FTC”) has issued an Advanced Notice of Proposed Rulemaking (“ANPR”) on “Commercial Surveillance and Data Security,”[1] initiating a proceeding intended to result in binding rules regarding “the collection, aggregation, analysis, retention, transfer, or monetization of consumer data and the direct derivatives of that information.”[2]

There is reason to believe that streamlined and uniform federal data-security or privacy regulations could be both beneficial and within the FTC’s competence and authority. But the approach suggested by the ANPR—simultaneously sweeping and vague—appears very likely to do more harm than good. Most notably, the ANPR evinces an approach that barely acknowledges either the limits of the FTC’s authority or the tremendous consumer benefits produced by the information economy.

The FTC is uniquely positioned to understand the complexities entailed in regulating privacy and data security. It has expertise and experience in both consumer-protection and competition matters. With regard to privacy and data security, in particular, it has decades of experience bringing enforcement actions for violations of the FTC Act’s prohibition of deceptive and unfair practices. Its enforcement experience also has been bolstered by its statutory mission to conduct economic and policy research, which has, not incidentally, comprised numerous hearings, workshops, studies, and reports on issues pertinent to data policy.

The ANPR does not build on the Commission’s experience and expertise as it could, however, and its dearth of economic analysis is especially striking. Moreover, the Commission’s authority is not unbounded, and neither are its resources. Both limitations are salient when the Commission considers adopting substantive—or “legislative”— regulations under either Section 18 or Section 6 of the FTC Act. As we discuss below, the current proceeding is deficient on both substantive and procedural grounds. Absent an express grant of authority and the requisite resources from Congress, the Commission would be ill-advised to consider, much less to adopt, the kinds of sweeping data regulations that the Commercial Surveillance ANPR appears to contemplate.

A.      The FTC Must Provide More Detail Than Is Contained in the ANPR

The ANPR states that it was issued pursuant to the Commission’s Section 18 authority,[3] which both grants and restrains the FTC’s authority to adopt regulations with respect to “unfair or deceptive acts or practices in or affecting competition” (“UDAP”).[4] Rulemaking under Section 18 of the FTC Act[5] requires that the Commission follow a careful process. As a preliminary matter, it must identify for both Congress and the public an area of inquiry under the Commission’s jurisdiction; the Commission’s objectives in the rulemaking; and regulatory alternatives under consideration.[6] Unfortunately, the Commission has not met these obligations in this ANPR.

Under Section 18, the Commission may adopt “rules which define with specificity acts or practices which are unfair or deceptive acts or practices in or affecting commerce”[7] under Section 5 of the FTC Act. Section 18 imposes express procedural requirements, in addition to those set out for this ANPR. These include, but are not limited to, requirements for a Notice of Proposed Rulemaking (“NPRM”). Section 18 also incorporates by reference the procedures prescribed by the Administrative Procedure Act.[8]

As noted, Section 18’s requirements for an ANPR are brief and preliminary but they are nonetheless real. In contravention of the requirements of Section 18, this ANPR does not clearly describe any “objectives which the Commission seeks to achieve,” and it provides no indication of “possible regulatory alternatives under consideration by the Commission.”[9] Instead, it provides a laundry list of putative harms, and it fails to identify even the most basic benefits that may be associated with diverse commercial-data practices. It does not describe the Commission’s current assessment of, or position on, those practices. And it provides no sense of the direction the Commission intends to take regarding potential rules.

Failing to identify the Commission’s objectives or proposals under consideration, this ANPR fails in its basic purpose to “invite… suggestions or alternative methods for achieving [the] objectives.”[10]

B.       The Commission Must Undertake a Cost-Benefit Analysis that Defines Harms, Identifies Benefits, and Weights the Two

Any rules the Commission issues under a Section 18 proceeding must emerge from a cost-benefit analysis.[11] Both the potential harms and the benefits of challenged conduct must be well-defined, and they must be weighed against each other. Even at this early stage of the process, the FTC is obligated to provide more than a suggestion that some harm might be occurring, and to provide more than a hint of how it might handle those harms.

This is also good procedure for policymaking more generally, irrespective of the Commission’s statutory obligations under Section 18. Before engaging in a deeply interventionist regulatory experiment—such as imposing strict privacy regulations that contravene revealed consumer preferences—the Commission should publicly state empirically justified reasons to do so. In other words, there should be demonstrable market failures in the provision of “privacy” (however we define that term) before centralized regulation co-opts the voluntary choices of consumers and firms in the economy, and before it supplants the ability to redress any residual, cognizable harms through law enforcement with broad, economywide, ex ante rules.

Thus, a vital threshold question for any rules issued under this proceeding is whether and why markets operating without specific privacy regulation generate a suboptimal provision of privacy protection. Without this inquiry, it is unclear whether there are problems requiring regulatory intervention and, if so, what they are. Without knowing their purpose, any rules adopted are likely to be ineffective, at best, and harmful, at worst. They may increase costs for consumers and businesses alike, chill innovation, mandate harmful prescriptions for alleged privacy harms while failing to address the most serious and persistent harms, or exacerbate the risks of harm—or all of the above.

Particularly in the United States, where informational privacy is treated both legally and socially as more of a consumer preference (albeit, perhaps, a particularly important one) than a fundamental right,[12] it is difficult to determine whether our current regime produces the “right” amount of privacy protection. That cannot be determined by observing that some advocates and consumers who are particularly privacy-sensitive opine that there should be more, or more of a certain sort; nor is it enough that there have been some well-publicized violations of privacy and cases of demonstrable harm. Indeed, the fact that revealed preferences in the market tend toward relatively less privacy protection is evidence that advocates may be seeking to create a level and a type of privacy protection for which there is simply no broad-based demand. Absent a pervasive defect that suggests a broad disconnect between revealed and actual preferences, as well as a pattern of substantial net harm, the Commission should be extremely cautious before adopting preemptive and sweeping regulations.

At a minimum, the foregoing indicates that the Commission must undertake several steps before this ANPR is close to satisfying the requirements of Section 18, not to mention good government:

• First, the Commission must proffer an adequate definition of “commercial surveillance.” While the ANPR is framed around this ominous-sounding term,^[13] it is functionally defined in a way that is both sweeping and vague. It appears to encompass virtually all commercial uses of “consumer data,” albeit without providing a workable definition of “consumer data.”[14] If the Commission is contemplating a general data regulation, it should say so and enumerate the objectives such a regulation would serve. In the current ANPR, the Commission has done neither.
• Second, the Commission must do more than merely cite diverse potential harms arising from what it terms “commercial surveillance.” The Commission has a long history of pursuing privacy and data-security cases, and it should rely on this past practice to define with specificity the types of harms—cognizable as injuries under Section 5—that it intends to pursue.

The Commission must also adequately account for the potential harms to innovation and competition that can arise from the adoption of new privacy and data-security regulations. Resources that firms invest in compliance cannot be invested in product development, customer service, or any of a host of other ends. And compliance with overly broad constraints will often curtail or deter the sort of experimentation that is at the heart of innovation.

Moreover, there is a potential tension between privacy and data security, such that mandates to increase privacy can diminish firms’ ability to ensure data security. The EU’s experience with the General Data Protection Regulation (“GDPR”) has demonstrated some of this dynamic.^[15] These realities must be incorporated into the Commission’s assessment.

• Third, the Commission must do more than merely nod to potential benefits that the modern data-driven economy provides to consumers. The clear benefits that arise from information sharing must be considered. Since the dawn of the Internet, free digital services have created significant consumer surplus. This trend continues today: Research using both survey and experimental methods has consistently found substantial benefits for consumers from sharing information in exchange for free (or subsidized) digital products. Moreover, productive conduct and consumer benefits are not limited to free digital products and services. Myriad products and services—from health care to finance to education—are made more efficient, and more widely available, by the commercial use of various forms of consumer data.

C.      The ANPR Must Account for the Effect of Any ‘Commercial Surveillance’ Rules on Consumer Welfare and Competition

The Commission is obligated to consider the likely effects of data regulation on consumers and competition. That ought to be a requirement for regulation generally, but it is an express, statutory requirement for unfairness regulation under Section 18 of the FTC Act. The Commission is uniquely well-situated to meet that mandate by virtue of its distinctive, dual competition and consumer-protection missions. Indeed, the Commission’s antitrust-enforcement experience dates to the agency’s inception. In addition, the Commission can access the considerable expertise of its Bureau of Economics, which employs experts in both industrial organization and consumer-protection economics. Yet much of that expertise appears absent from the ANPR.

This ANPR does not specify, or even sketch, the data regulations being contemplated by the Commission. Neither does it specify the Commission’s goals in the rulemaking or alternative regulatory approaches under consideration, although both are required by statute. Consequently, one cannot assess the net effects of any proposed “commercial surveillance and data security” rule on competition or consumers, because there simply is no proposed rule to assess.

The economic literature, however, does suggest caution:

• First, as a general matter, regulations that impose substantial fixed costs on regulated firms tend to burden smaller firms and entrants more than they do large firms and incumbents.[16]
• Second, studies of specific domestic-privacy and data-security requirements underscore the potential for unintended consequences, including competitive costs.[17]
• Third, empirical studies of the effects of general data regulations in foreign jurisdictions, such as the EU’s GDPR, suggest that such regulations have indeed led to substantial competitive harms.[18]

The literature on the effects of GDPR and other data regulations is particularly instructive. Although it is neither definitive nor complete, it has thus far found slender (at best) benefits to competition or consumers from data regulations and considerable costs and harms from their imposition. Further experience with and study of data regulations could yield a more nuanced picture. And, again, the FTC is well-positioned to contribute to and foster a greater understanding of the competitive effects of various types of data regulation. Doing so could be greatly beneficial to policymaking, competition, and consumer welfare, precisely because specific data practices can produce substantial benefits, harms, or a complex admixture of the two. But documented harms and speculative benefits of regulation recommend caution, not blind intervention.

D.      Conclusion

The Commission should take account of a further reality: the rules it contemplates will be created in an environment filled with other privacy regulators. Although the United States does not have a single, omnibus, privacy regulation, this does not mean that the country does not have “privacy law.” Indeed, generally applicable laws providing a wide range of privacy and data-security protections already exist at both the federal and state level. These include consumer-protection laws that apply to companies’ data use and security practices,[19] as well as those that have been developed in common law (property, contract, and tort) and criminal codes.[20] In addition, there are sector-specific regulations pertaining to particular kinds of information, such as medical records, personal information collected online from children, and credit reporting, as well as regulations prohibiting the use of data in a manner that might lead to certain kinds of illegal discrimination.[21]

Despite the FTC’s noted experience in a certain slice of privacy regulation, Congress has not made the FTC the central privacy regulatory body. Neither has Congress granted the Commission the resources likely required for such a regulator. Congress has wrestled with complex tradeoffs in several areas and has allowed—through design and otherwise—various authorities to emerge. Where Congress has provided for privacy regulation, it has tailored the law to address specific concerns in specific sectors, or with respect to specific types of information. Moreover, in each case, it has balanced privacy and security concerns with other policy priorities. That balancing requires technical expertise, but it also entails essentially political judgements about the relative value of diverse policy goals; in that latter regard, it is a job for Congress.

There are, as well, questions of resource allocation that may attend an express statutory charge. We cannot gainsay the importance of the FTC’s privacy and data-security enforcement work under Section 5 of the FTC Act. At the same time, we cannot help but notice a misfit between the Commission’s congressionally allocated resources and the obligations that are entailed by data regulations of the scope contemplated in the ANPR. By way of contrast, we note that, since the compliance date of the Health Insurance Portability and Accountability Act (“HIPAA”) privacy rule, the U.S. Department of Health and Human Services (“HHS”) Office of Civil Rights (“OCR”) has investigated and resolved nearly 30,000 cases involving HIPAA-covered entities and their business associates; for appropriate cases of knowing disclosure or obtaining of protected health information, OCR has referred more than 1,500 cases to the U.S. Department of Justice (“DOJ”) for criminal prosecution.[22]

In his dissent from the issuance of this ANPR, former Commissioner Noah Phillips noted the massive and complicated undertaking it initiates:

Legislating comprehensive national rules for consumer data privacy and security is a complicated undertaking. Any law our nation adopts will have vast economic significance. It will impact many thousands of companies, millions of citizens, and billions upon billions of dollars in commerce. It will involve real trade-offs between, for example, innovation, jobs, and economic growth on the one hand and protection from privacy harms on the other. (It will also require some level of social consensus about which harms the law can and should address.) Like most regulations, comprehensive rules for data privacy and security will likely displace some amount of competition. Reducing the ability of companies to use data about consumers, which today facilitates the provision of free services, may result in higher prices—an effect that policymakers would be remiss not to consider in our current inflationary environment.^[23]

This is particularly true given the Commission’s long history of work in this area. The Commission has undertaken decades of investigations and a multitude of workshops and hearings on privacy and related topics. This ANPR nods to that history, but it does not appear to make much use of it, possibly because much of it contains lessons that pull in different directions. Overall, that impressive body of work does not remotely point to the need for a single, comprehensive privacy rule. Rather, it has demonstrated that privacy regulation is complicated. It is complicated not just as a technical matter, but also because of the immense variety of consumers’ attitudes, expectations, and preferences with respect to privacy and the use of data in the economy.

The Commercial Surveillance ANPR poses 95 questions, many of which will find some answers in this prior history if it is adequately consulted. The Commission has generally evidenced admirable restraint and assessed the relevant tradeoffs, recognizing that the authorized collection and use of consumer information by companies confers enormous benefits, even as it entails some risks. Indeed, the overwhelming conclusion of decades of intense scrutiny is that the application of ex ante privacy principles across industries is a fraught exercise, as each industry—indeed each firm within an industry—faces a different set of consumer expectations about its provision of innovative services and offering of privacy protections.

These considerations all militate in favor of regulatory restraint by the FTC as a matter of policy. They also require restraint, and an emphasis on established jurisdiction, given the Supreme Court’s recent “major questions” jurisprudence.[24] As noted in the statements of several commissioners, West Virginia v. EPA[25] clarifies the constitutional limits on an agency’s authority to extend the reach of its jurisdiction via regulation. In brief, the broader the economic and political sweep of data regulations the Commission might propose, the more likely it is that such regulations exceed the FTC’s authority. If the “major questions doctrine” is implicated, the burden is on the agency to establish the specific grant of authority that is claimed.[26] Moreover, the Court was clear that a merely colorable claim of statutory implementation is inadequate to establish the authority to issue sweeping regulations with major economic and political implications.[27]

Download the full comments here.

 

[1] Trade Regulation Rule on Commercial Surveillance and Data Security, 87 Fed. Reg. 51273 (Aug. 22, 2022) (to be codified at 16 C.F.R. Ch. 1) [hereinafter “ANPR” or “Commercial Surveillance ANPR”].

[2] Id. at 51277.

[3] Id. at 51276.

[4] That is, “unfair or deceptive acts or practices in or affecting commerce,” as they are prohibited under Section 5 of the FTC Act, 15 U.S.C. § 45(a)(1).

[5] 15 U.S.C. § 57a.

[6] 15 U.S.C. § 57a(b)(2)(A).

[7] 15 U.S.C. § 57a(a)(1)(B).

[8] 15 U.S.C. § 57a(b)(1) (“When prescribing a rule under subsection (a)(1)(B) of this section, the Commission shall proceed in accordance with section 553 of title 5.”)

[9] 15 U.S.C. § 57a(b)(2)(i).

[10] 15 U.S.C. § 57a(b)(2)(ii).

[11] See Section III, infra (regarding the role of cost-benefit analysis under Magnuson-Moss and the statutory requirements of Section 18).

[12] Except, of course, when it comes to government access to private information, i.e., under the Fourth Amendment.

[13] See, e.g., ANPR, supra note 1 at 51273-75.

[14] The purported definition of consumer data in the ANPR, and the scope of activities around consumer data, are so overbroad as to encompass virtually the entirety of modern economic activity: “the collection, aggregation, analysis, retention, transfer, or monetization of consumer data and the direct derivatives of that information. These data include both information that consumers actively provide—say, when they affirmatively register for a service or make a purchase—as well as personal identifiers and other information that companies collect, for example, when a consumer casually browses the web or opens an app. This latter category is far broader than the first.” Id. at 51277.

[15] See, e.g., Coline Boniface, et al., Security Analysis of Subject Access Request Procedures, in Privacy Technologies & Policy: 7^th Annual Privacy Forum (Maurizio Naldi, et al. eds., 2019).

[16] See, e.g., James Campbell, Avi Goldfarb & Catherine Tucker, Privacy Regulation and Market Structure, 24 J. Econ. & Mgmt. Strategy 47 (2015); Alex Marthews & Catherine Tucker, Privacy Policy and Competition, Econ. Stud. at Brookings (December 2019), available at https://www.brookings.edu/wp-content/uploads/2019/12/ES-12.04.19-Marthews-Tucker.pdf.

[17] See, e.g., Jin-Hyuk Kim & Liad Wagman, Screening Incentives and Privacy Protection in Financial Markets: A Theoretical and Empirical Analysis, 46 RAND J. Econ. 1 (2015).

[18] See, e.g., Jian Jia, Ginger Zhe Jin & Liad Wagman, The Short-run Effects of the General Data Protection Regulation on Technology Venture Investment, 40 Marketing Sci. 661 (2021).

[19] See, e.g., FTC Act, 15 U.S.C. § 45(a) et seq.

[20] See Privacy-Common Law, Law Library —American Law and Legal Information, http://law.jrank.org/pages/9409/Privacy-Common-Law.html (last visited Oct. 16, 2022).

[21] See, e.g., Comments of the Association of National Advertisers on the Competition and Consumer Protection in the 21^st Century Hearings, Project Number P181201, available at https://docplayer.net/93116976-Before-the-federal-trade-commission-washington-d-c-comments-of-the-association-of-national-advertisers-on-the.html: [T]he Health Information Portability and Accountability Act (“HIPAA”) regulates certain health data; the Fair Credit Reporting Act (“FCRA”) regulates the use of consumer data for eligibility purposes; the Children’s Online Privacy Protection Act (“COPPA”) addresses personal information collected online from children; and the Gramm–Leach–Bliley Act (“GLBA”) focuses on consumers’ financial privacy; the Equal Employment Opportunity Commission (“EEOC”) enforces a variety of anti-discrimination laws in the workplace including the Pregnancy Discrimination Act (“PDA”) and American with Disabilities Act (“ADA”); the Fair Housing Act (“FHA”) protects against discrimination in housing; and the Equal Credit Opportunity Act (“ECOA”) protects against discrimination in mortgage and other forms of lending. Id. at 6.

[22] Dep’t Health & Human Servs., Health Information Privacy, Enforcement Highlights, https://www.hhs.gov/hipaa/for-professionals/compliance-enforcement/data/enforcement-highlights/index.html (HHS Office of Civil Rights, last reviewed Sep. 14, 2022).

[23] ANPR at 51293 (Dissenting Statement of Comm’r Noah J. Phillips).

[24] See W. Virginia v. Env’t Prot. Agency, 142 S. Ct. 2587, 2595 (2022) (citing a line of cases including Utility Air Regulatory Group v. EPA, 573 U. S. 302 (2014); Gonzales v. Oregon, 546 U. S. 243 (2006); FDA v. Whitman v. American Trucking Assns., Inc., 531 U. S. 457, 468 (2001); and Brown & Williamson Tobacco Corp., 529 U. S. 120, 159 (2000)).

[25] Id.

[26] See id. at 2613 (citing William Eskridge, Interpreting Law: A Primer on How to Read Statutes and the Constitution 288 (2016)).

[27] Id. at 2608-09.

Executive Summary

For years, regulators and competition watchdogs have expressed concern about competition in the digital advertising business. They note that digital advertising appears to be dominated by a few dominant firms, such as Google, Facebook, and—to a lesser extent—Amazon. Some claim that this dominance allows these firms—and Google, in particular—to engage in anticompetitive conduct to extend their market power and to earn supercompetitive profit at the expense of advertisers, publishers, and consumers. This paper investigates the digital advertising market and assesses some of these claims. We conclude based on the information that is publicly available that many of the most significant claims made against Google’s advertising technology (ad tech) business are based on a misunderstanding of U.S. antitrust law, or of the details of the  ad tech market itself.

Digital advertising provides the economic underpinning for much of the Internet. Targeted digital advertising on independent websites is often facilitated by intermediaries that match advertisers and websites automatically, displaying ads to users for whom they are most relevant. This intermediation has advanced enormously over the past three decades. Some now allege, however, that the digital advertising market is monopolized by its largest participant: Google.

In particular, a lawsuit filed in December 2020 by the State of Texas and nine other U.S. states (later joined by five more states in March 2021) alleges anticompetitive conduct related to Google’s online display-advertising business. It has been reported that the U.S. Justice Department (DOJ) may bring a similar lawsuit before the end of 2022.

Moreover, a bipartisan group of U.S. senators introduced the Competition and Transparency in Digital Advertising Act in May 2022. If passed by Congress and signed into law, the bill would require some of the largest Internet firms—such as Google, Facebook, and Amazon—to break up their digital advertising businesses. A summary of the bill claims that Google is the “leading or dominant” firm in “every part” of the ad tech “stack” and that it uses this dominance to “extract monopoly rents” from advertisers and publishers.

Our paper focuses on the open-display digital advertising business. Display ads are designed to be visually engaging, combining text, images, and a hyperlink to a website. These are distinct from search ads, which are text ads displayed along with organic Internet search results. Most of today’s digital display advertising appears on heavily trafficked owned-and-operated sites such as TikTok, Instagram, YouTube, and Amazon, in which the company providing the advertising space is the same company that operates the platforms that place the ads. In contrast, open-display space is supplied by independent publishers—such as The New York Times (nytimes.com), MLB.com, or The Weather Channel (weather.com)—and is usually facilitated by intermediaries.

This paper begins with an overview of digital advertising. The market’s history is one of dynamic innovation, with many new developments arising to solve problems created by previous innovations, address new innovations, and respond to market developments. The market’s structure has changed dramatically as advertisers, publishers, and consumers have responded to new technologies. These changes and innovations all must balance the competing demands of advertisers, publishers, and consumers to maximize the total value of the advertising platform. Thus, any antitrust evaluation of digital advertising must consider whether the market’s structure and the conduct of its participants may be procompetitive responses to prior market changes, as well as the extent to which the overall market structure may mitigate superficially problematic elements within it.

Of particular importance, digital advertising intermediaries that are vertically integrated into some or all components of the digital advertising “stack” of services use the prices charged to each side of the market to optimize overall use of the platform. As a result, pricing in these markets operates differently than pricing in traditional markets: pricing on one side of the platform is often used to subsidize participation on another side of the market, increasing the value to all sides combined. Consequently, pricing (or other terms of exchange) on one side of the market may appear to diverge from the competitive level when viewed for that side alone. While one side of the market may pay superficially higher fees, this cost can be offset by the benefits from increased participation on the other side of the market. In this way, using subsidies to increase participation on another side of the market creates valuable network effects for the side of the market facing the higher fees.

In the second half of the paper, we address some of the specific allegations of anticompetitive structures and conduct made in the Texas Complaint and by critics of the digital advertising industry. We conclude that a flawed premise underlies many of these allegations. It is a version of the “big is bad” argument, in which conduct by dominant incumbent firms that makes competition more difficult for certain competitors is viewed as inherently anticompetitive—even if the conduct confers benefits on users. Under this approach, the largest firms are seen as acting anticompetitively if they do not share their innovations or reveal their business processes to competing firms. As a result, creating new and innovative products, lowering prices, reducing costs through vertical integration, and enhancing interoperability among existing products is miscast as anticompetitive conduct.

In contrast, we note that U.S. antitrust law is intended to foster innovation that creates benefits for consumers, including innovation by incumbents. The law does not proscribe efficiency-enhancing unilateral conduct on the grounds that it might also inconvenience competitors, or that there is some other arrangement that could be “even more” competitive.

Moreover, U.S. antitrust law does not second guess unilateral conduct simply because it may hinder rivals. Any such conduct would first have to be shown to be anticompetitive—that is, to harm consumers or competition, not merely certain competitors. In multisided markets, this means finding not simply that some firms on one side of the market are harmed, but that the combined net effect of challenged conduct across all sides of the market is harmful. The Texas Complaint, however, is built on the alleged harm of reduced revenue to publishers, without considering the corresponding benefit of lower prices to advertisers (and the consumers of advertised products and services).

Based on the information publicly available, we conclude that many of the most significant claims made against Google’s ad tech business are based on a misunderstanding of U.S. antitrust law, or of the details of the ad tech market itself. Although we cannot be sure how the Texas et al v. Google case will develop once its allegations are fleshed out into full arguments, many of its claims and assumptions appear wrongheaded. If the court rules in favor of these, the result will be to condemn procompetitive conduct and potentially to impose costly, inefficient remedies that function as a drag on innovation.

Legislators, too, who may be concerned about Google’s conduct and tempted to impose regulatory requirements on tech companies should bear in minds the risk of the “Nirvana fallacy,” in which real-life conduct is compared against a hypothetical “competition-maximizing” benchmark and anything that falls short is deemed worthy of intervention. That fanciful approach would pervert businesses’ incentives to innovate and compete and would make an unobtainable “perfect” that exists only in the minds of some economists and lawyers the enemy of a “good” that exists in the market.

Introduction

For years, regulators and competition watchdogs have expressed concern about competition in the digital advertising business. They note that digital advertising appears to be dominated by a few exceptionally large firms, such as Google, Facebook, and—to a lesser extent—Amazon. Some claim that this dominance allows these firms—and Google, in particular—to engage in anticompetitive conduct to extend their market power and to earn supercompetitive profits at the expense of advertisers, publishers, and consumers. This paper investigates the digital advertising market and assesses some of these claims. We conclude, based on the information that is publicly available, that many of the most significant claims made against Google’s advertising technology (ad tech) business are based on a misunderstanding of U.S. antitrust law, or of the details of the ad tech market itself.

Digital advertising provides the economic backbone for much of the Internet. By providing websites and apps a means to monetize their products without having to charge user fees, advertising enables access to entertaining and informative content without payment. Targeted advertising allows companies to inform potential customers of new products, giving new entrants a way to compete with popular incumbents, while effective targeting avoids wasting the time of those who aren’t likely to be interested. Advertising can endow products with new characteristics in customers’ minds and make consumers aware of product features they may not have known about.

Advertising on independent websites is often facilitated by intermediaries that match advertisers and websites automatically, targeting ads at the users to whom they are most relevant. This intermediation has advanced enormously over the past three decades. Some now allege, however, that the digital-advertising market is monopolized by its largest participant: Google. These allegations originate from various sources, including policy discussions, lawsuits, draft legislation, and academic reports.

In particular, a lawsuit filed in December 2020 by the State of Texas and nine other U.S. states (later joined by five more states in March 2021) alleges, among other things, anticompetitive conduct related to Google’s online display-advertising business.[1] This action is one of three currently pending lawsuits brought against Google by government antitrust enforcers in the United States; the other two relate to Google’s distribution agreements and search design.[2] It has been reported that the U.S. Department of Justice (DOJ) may bring a similar lawsuit before the end of 2022.[3] Along similar lines, the European Commission opened an investigation into Google’s display-advertising services in June 2021[4] and the German competition authority has published a report regarding its inquiry into non-search advertising.[5] Among other things, the European Commission is investigating whether Google “has made it harder for rival online advertising services to compete in the so-called ad tech stack.”[6]

These ongoing cases follow regulatory reports and hearings examining the market, including a year-long study by the United Kingdom’s Competition and Markets Authority (CMA). The CMA investigation of digital advertising (including search and social-media advertising) has thus far produced recommendations for a code of conduct and “pro-competitive interventions” into the market, as well as a new regulatory body to oversee these measures.[7] The Australian Competition and Consumer Commission is also conducting its own study of the digital advertising market,[8] and both houses of the U.S. Congress have held hearings on the market in recent years.[9] In October 2020, the Democratic majority staff of the U.S. House Judiciary Committee’s Subcommittee on Antitrust, Commercial, and Administrative Law issued a report that recommended, among other things, regulation for the display advertising market.[10]

The digital  advertising industry has also drawn the attention of legislators. In May 2022, a bipartisan group of U.S. senators introduced a bill that would require some of the largest Internet firms to break up their digital advertising businesses.[11] Dubbed the Competition and Transparency in Digital Advertising Act (CTDAA), the measure was introduced by Sen. Mike Lee (R-Utah) and co-sponsored by Sens. Amy Klobuchar (D-Minn.), Ted Cruz (R-Texas), and Richard Blumenthal (D-Conn.). Sen. Lee’s summary of the CTDAA identifies several allegations against the largest firms in the digital advertising business, with an emphasis on Google.[12] The summary claims that Google is the “leading or dominant” firm in “every part” of the “ad tech stack” and that it “exploits” conflicts of interest to “extract monopoly rents” from advertisers and publishers. Because of these monopoly rents, consumers are harmed in the form of higher prices for advertised goods and services and lower quality of online content, according to Lee’s press release.[13]

A 2020 paper published by the Omidyar Network—itself based on an interim CMA report produced during the authority’s then-pending investigation[14]—alleged anticompetitive practices within Google’s display advertising business and laid out a “roadmap” for a prospective antitrust case.[15] Other legal and economic consultants have also voiced concerns about Google’s role in the digital advertising industry.[16] These critiques were published before the Texas Complaint and provide more detail underlying the allegations and arguments described in the Texas Complaint. For that reason and because it may influence further lawsuits and regulatory interventions in the digital advertising market, including the DOJ’s, we also evaluate several of the Roadmap’s findings and conclusions.

This paper investigates the digital advertising market and assesses the aforementioned claims that it is uncompetitive. It explains some of the complex dynamics that underpin this market, thereby shedding light on the weaknesses and deficiencies in many of the arguments made about it, particularly those behind the Texas Complaint. This analysis is relevant to the entire Internet and to the wider economy, not just to Google and the display-advertising market. Many of the allegations made in the Texas Complaint would, if upheld by a court, have profound implications for antitrust law by establishing precedents that successful platforms effectively have a legal duty to act as essential facilities for their competitors; that efficiency-enhancing innovation by incumbent platforms is anticompetitive (particularly when it is not shared with competitors); and that courts or regulators can impose remedies that put these duties into effect without consideration of the harmful tradeoffs and unforeseen consequences that could themselves constrain competition and innovation. Such an approach would severely affect not only Google and the ad tech industry, but also businesses operating in unrelated markets and industries.

We begin with an overview and history of digital advertising. It is a history of dynamic innovation, with many new developments arising to solve problems created by previous innovations, address new innovations, and respond to market developments. These innovations must balance the competing demands of advertisers and publishers. The market’s structure has changed dramatically as advertisers, publishers, and consumers have responded to recent technologies. Because of this dynamism, we argue that it is a mistake to conclude that market structure and firm conduct at some specific point in time was ideal or better from a competition point of view, or that deviations from that paradigm represent a problem for competition enforcers to correct.

In the second half of the paper, we address some of the specific allegations of anticompetitive structures and conduct made in the Texas Complaint and by critics of the digital advertising industry. We conclude that a flawed premise underlies many of these allegations. Fundamentally, the allegations stem from an assertion that conduct engaged in by dominant, incumbent firms that makes competition more difficult for competitors is anticompetitive—even if the conduct confers benefits on users. This often amounts to a claim that the largest firms are acting anticompetitively by innovating and developing their business processes and products in ways that create benefits for one or more digital advertising constituents and for the advertising ecosystem more generally, including by creating new and innovative products, lowering prices, reducing costs through vertical integration, and enhancing interoperability between existing products.

I. Overview of the Digital Advertising Industry

For many people, digital ads are “just there.” They appear on one’s Facebook timeline, get inserted into one’s Twitter feed, or show up in the middle of an online news article. Unseen to most users is a complex stack of services that match advertisers with advertising space, using real-time auctions and other algorithms to deliver ads targeted to produce a user response, such as buying a product, supporting a cause, or visiting another website.

In this section we explain that digital advertising is just one small part of a much broader advertising and marketing industry. We provide a concise history of how the business evolved from simple banner ads to highly targeted display ads. Through this evolution, digital advertising has become a multisided market where intermediaries must balance the demands of advertisers, publishers, and users to maximize the total value of the advertising platform. Because of this balancing, it is a mistake for policymakers and regulators to focus only on a single set of users or a single segment of the stack of digital advertising services.

A.   Digital advertising is part of a broader advertising and marketing market

The Texas Complaint alleges that Google has market power in six distinct product markets, each of which the states claim to be nationwide in geographic scope. Four of these distinct product markets involve intermediation in the sale of “open” display ads on third-party websites and two involve intermediation in the sale of “in-app” display ads on mobile devices.[17] As we note in our earlier paper, it is likely that the states allege overly narrow product-market definitions.[18] In particular, structure and conduct viewed within a broader digital advertising market, overall advertising market, or marketing market indicates than no single firm has significant market power.

Digital advertising comprises about half of U.S. advertising revenues (Figure 1), while advertising itself accounts for about half of marketing activities. Marketing includes advertising, as well as events, sales promotion, direct marketing, telemarketing, product placement, and merchandising. Within digital advertising, advertisers have a broad set of options about where and how to run ads, including:

• Search ads, in which the ad is displayed as a search-engine result (e.g., Google, Bing, DuckDuckGo);
• Display ads on a site owned and operated by the firm that sells the ad space (e.g., Facebook, YouTube, Amazon Marketplace);
• “Open” display ads on a third party’s site (e.g., The New York Times, Dallas Morning News, Runner’s World); or
• “In-app” display ads served on mobile apps.

While total advertising spending in the United States has increased by about 15% since 2009, as a share of the economy spending has been relatively flat at slightly less than 1% of GDP.[19] As mentioned, about half of total U.S. advertising dollars go to digital channels, up from approximately 10% a decade ago. Approximately 30% of ad spending goes to TV and less than one-quarter goes to radio, newspapers, magazines, billboards, and other “offline” forms of media.[20]

Figure 1: U.S. Advertising Spending Over Time

Source: Benedict Evans, News by the Ton

It is well-understood that television broadcasters and cable networks compete with digital services.[21] And they do so on virtually all dimensions: for user attention, for labor, for content and other inputs—and for advertising. The same is true of competition for advertising among digital publications, newspapers, radio, magazines, video games, music streaming, and podcasts. The fact that offline and online advertising—to say nothing of marketing more broadly—employ distinct processes does not consign them to separate markets. Indeed, it is widely accepted that online advertising has drawn advertisers from offline markets, as previous technological innovations drew advertisers from other channels before them.[22] Moreover, while there is evidence that, in some cases, offline and online advertising may be complements as well as substitutes,[23] the distinction between these cases is becoming less and less meaningful as the revolution in measuring the effectiveness of advertising has changed how marketers approach different levels of what is known as the marketing “funnel.”[24] For example, economist David Evans’ review of the literature concludes that digital advertising is a segment of the broader advertising business in which different forms of advertising compete and complement each other:

Advertisers base decisions about the level and allocation of their budgets on formal or informal analyses of the rate of return on investment. For these ad campaigns, the different advertising methods can be substitutes to the extent they provide alternative ways of delivering messages to an audience, and complements to the extent they can reinforce each other.[25]

Economists Avi Goldfarb and Catherine Tucker demonstrate that display-advertising pricing is sensitive to the availability of offline alternatives.[26] Although advertising technology and both supplier and consumer preferences continue to evolve, the weight of evidence suggests a far more unified and integrated economically relevant market between offline and online advertising than their common semantic separation would suggest:

We believe our studies refute the hypothesis that online and offline advertising markets operate independently and suggest a default position of substitution. Online and offline advertising markets appear to be closely related. That said, it is important not to draw any firm conclusions based on historical behavior.[27]

In summary, there is evidence that open-display and in-app ads compete with search ads, while digital ads compete with offline advertising. Thus, courts and regulators should be skeptical of overly narrow market definitions focused on only small slices of a much larger relevant market for advertising.

B.   A simplified description of digital display advertising

The combination of software and processes that facilitate digital advertising transactions is known as the “ad tech stack.” The stack consists of several software components to match advertisers with publishers.

Advertiser ad servers are used by advertisers and media agencies to store ads, deliver them to publishers, track their activity, and assess their effectiveness (by, for example, tracking conversions). Demand-side platforms (DSPs) automate the purchase of advertising inventory by collecting bids in real-time auctions from multiple advertiser ad servers. DSP bids are based on the advertiser’s objectives, data about the final user, and data on impressions or conversions. Publisher ad servers manage publishers’ inventory and determine whether and where to serve a particular ad on a publisher’s site. Supply-side platforms (SSPs) automate the sale of publishers’ inventory, typically through real-time auctions involving multiple DSPs.

In general, the process of buying and selling digital ads through open-display auctions works as follows (Figure 2):

1. When a user opens a webpage (or uses an app), the publisher’s ad server sends a bid request to SSPs for the advertising impressions available on that page for that user.
2. The SSPs send bid requests to multiple DSPs for the ad impressions.
3. DSPs evaluate the advertising opportunity based on user data and the objectives of their advertisers’ campaigns and send bids to the SSPs.
4. SSPs rank the bids received based on price and other priorities set by the publisher. The SSPs send their winning bids to the publisher.
5. The publisher ad server compares bids received from SSPs, along with any pre-existing direct deals between the publisher and specific advertisers and decides which ad to serve on the page.

Figure 2: A Simplified View of the Digital Advertising Stack

Source: OECD, CMA

While this process applies to most programmatic transactions, there are many variations. For example, there are diverse ways in which SSPs are contacted and asked to submit their bids. To the extent that a publisher and advertiser have a pre-existing, direct agreement, there are differences in how these arrangements are handled and integrated with deals arranged through automated platforms. The specific approach used to match ads with ad inventory will reflect a balance among different sides of a multisided market. One approach might increase the prices received by sellers (publishers) but expose buyers (advertisers) to increased risk of overpayment. Other methods might reduce risks to advertisers, but also reduce the prices received by publishers.

C.   A brief history of digital advertising

This history of the digital advertising market is a history of iterative innovation, with new developments and services arising to solve problems created by previous innovations and to respond to changing market conditions. At the heart of these innovations is an attempt to balance the competing demands of advertisers, publishers, and consumers. Given that this is a dynamic market, it would be mistaken to conclude that the market structure at some specific point was the “correct” one from a competition point of view. Moreover, it would be a mistake to conclude that deviations from some previous “ideal” world present a problem that can be corrected by disruptive regulation.

Digital advertising originally worked similarly to conventional print and broadcast advertising. Online publishers would negotiate with advertisers (or their ad agencies) to sell ad space on their websites, giving relevant advertisers information about their readership gathered through market research. All users would see the same ads. The resulting system was poorly targeted, inefficient, and carried high fixed costs, including the cost of things like market research and the transaction costs of publishers hiring sales teams and advertisers hiring ad agencies to do business with one another. Inevitably, these fixed costs meant that only larger publishers and larger advertisers could engage in the online market profitably.

In 1993, O’Reilly & Associates Inc. introduced its Global Network Navigator (GNN) magazine and other ad-supported online publications, which first rolled out clickable ads. O’Reilly is credited with the first attempt to create an “advertising medium” on the Internet.[28] The price of ads ranged from $500 for a one-page business profile to $5,000 for up to 25 pages about the company placing the ad.[29] A year later, Wired magazine’s digital affiliate HotWired ran what later became known as the web’s first banner ads. The ad—for AT&T’s “You Will” campaign—cost $30,000 for a three-month dedicated placement in a section of HotWired’s site.[30]

The first step toward automating this process came with the introduction of ad-server software on both the publisher and advertiser sides, which allowed each side to automate parts of the ad-placement process. Ad servers allowed publishers to automatically describe the type of content on their pages, which in turn allowed advertisers to place ads tailored to that content. An article about hiking could automatically indicate to a department store to place ads selling walking boots, for instance. It also allowed the publisher to sell to many advertisers without having to transact directly with any of them. Ad servers also allowed advertisers to browse and manage campaigns across a large, aggregated number of publishing sites, instead of having to interact with sales teams one by one. This process was, however, still negotiated directly, and often left publishers with unused “remnant” advertising space that they had not been able to sell.

To solve this problem, ad networks entered the market. These functioned as intermediaries between advertisers’ and publishers’ ad servers, aggregating unsold remnant ad space and allowing advertisers to buy that space en masse without having to deal directly with each publisher. Ad networks did not replace direct sales, but they allowed for residual space to be bought and sold more easily, increasing the amount of ad inventory available and lowering the fixed costs to use it. This, in turn, made the market feasible for smaller publishers, who would otherwise be unable to attract direct deals to sell ad space, and for advertisers to conduct large-scale ad campaigns across many publishers (including small ones).

In 1995, GNN was acquired by AOL.[31] That same year, marketing-communications agency Poppe Tyson spun off its Internet advertising division as DoubleClick, with the objective of “responding to advertisers’ need to be able to buy millions of impressions on the Internet without having to buy from hundreds of different sites.”[32] The company created “subnets” of publishers to target specific categories of consumers.[33]

Also in 1995, ad agency WebConnect, the first ad network, began to collaborate with its clients to identify the websites that their ideal consumers visit. WebConnect then placed ads on the websites where they were more likely to be seen by the audience most relevant to their clients. The company also produced a tool to prevent “ad fatigue,” which occurs when users are repeatedly shown the same ad.[34]

In 1998, GoTo.com, which was renamed Overture in 2001, launched its ad-supported search engine.[35] Search result rankings were based on an open-market bidding process. Advertisers on GoTo were informed of the amounts other advertisers were bidding for a click-through within the results for a given search term, and any advertiser could increase its bid to obtain a higher ranking, a process GoTo described as “pay for performance.”[36] One of GoTo’s key innovations was linking advertising pricing to click-throughs, rather than to page views.[37]

To drive home just how efficient these ads could be, [GoTo.com’s founder Bill Gross] came up with an audacious pricing scheme. Instead of paying for page-views—an old-media model that had come to dominate the Web—advertisers would pay only when people actually clicked on their ads. And their placement on the GoTo.com results page would be determined through an auction, so that more desirable keywords would command higher prices, while less common keywords could be had for as little as a penny per click. As a search engine, GoTo.com had nothing on Google. But as a way of making money on searches, it was ingenious.[38]

During the dot-com boom of the late 1990s, banner ads spread throughout the Internet, though growth was tempered by user complaints that the ads slowed page loading.[39] The dotcom bust wiped out many of the firms that were the biggest buyers of digital banner ads. In response, Wired predicted that digital advertising would undergo a “facelift.”[40]

Old media revenues will wither as mainstream advertisers storm the Net. Instead of stuffing junk mail into the mailbox outside your house, they’ll send it directly to your inbox. And companies get smarter, choosing sites that take better aim at their quarry, er, potential customers.

“It’s very much a targeted medium,” Robin Neilfield, co-founder of NetPlus Marketing. “You have to analyze the content on a site, you can’t just buy based on demographics.”[41]

Because ad networks were not comprehensive—they did not carry the entire inventory of the Internet—publishers began to use yield managers (later known as SSPs) to compare bids for their ad space and to decide which to accept. SSPs effectively allowed publishers to aggregate demand from a larger number of ad networks, which themselves aggregated demand from advertisers. This process allowed ad space to be more easily commoditized, with an SSP helping to identify an ad space’s relevance to potential advertisers.

As indirect sales became possible, ad exchanges emerged to sell ad space using real-time auctions. Ad space could be tagged according to characteristics like audience type, relevance to the advertiser, and/or prominence and quality of the ad, with bids gauged accordingly.

Finally, DSPs on the advertiser side allowed advertisers to engage with many ad networks and ad exchanges at one time. These also allowed advertisers to track campaigns and measure performance of different ads with different publishers, and to adjust their campaigns accordingly. Most ad exchanges now have DSP functionality built in.

In 2000, Google introduced a new self-service advertising product called AdWords (now Google Ads) that allowed businesses to purchase text ads on search-results pages. At the time it was reported that AdWords was designed to attract small-to-midsize advertisers with budgets of $5,000 or less.[42] AdWords differed from GoTo/Overture in a major way. GoTo/Overture placed ads within search results, with results ranked by bid. In contrast, AdWords placed ads separate from search results with pricing based on pageviews.[43] In this way, Google could display ads without compromising the relevance of search results. In 2002, it launched AdWords Select, its pay-per-click, auction-based search-advertising product.[44]

In 2003, Google acquired Applied Semantics, whose AdSense display advertising product allowed it to sell targeted ads on third-party websites.[45] With AdSense, the display-ad server was able to read text on a publisher’s site and serve relevant ads, considering factors like the user’s geographic location, age, demographics, and the search made.[46] AdSense was the forerunner of programmatic display advertising, the process of automating the buying and selling of ad inventory in real time through an automated bidding system. In 2005, Google introduced the Quality Score model, which considers an ad’s click-through rate, as well as the bid price, in placing ads.[47]

YouTube was launched in 2005 and acquired by Google the following year, when the company also introduced video ads. In 2007, Google acquired DoubleClick.[48]

Around 2015, “header bidding” began to roll out, with publishers Meredith Corp. and Townhall Media as two of largest early adopters.[49] Before header bidding, it was difficult for every demand-side partner to submit a bid for every ad request. As a result, publishers relied on approaches such as “ad waterfalls”[50] to try to get the most from each partner. Because of the way ad waterfalls are configured (based on historical, not real-time, data), publishers believed ad waterfalling led to winning bids that were below what some bidders might be willing to pay.[51] Client-side[52] header bidding was adopted as a way to increase real-time price competition among multiple SSPs, leading to higher returns for publishers and a more efficient allocation of ad space to advertisers.

Despite the widespread adoption of header bidding—as of the second quarter of 2021, about two-thirds of publishers were using it[53]—the technology has its own challenges. For example, the addition of extra code on the webpage, which client-side header bidding requires, can slow down the publisher’s website, driving away users.[54]

As an alternative to client-side header bidding, server-side header bidding was introduced. Prebid launched in 2015 as an independent and open-source option. Google released Open Bidding in April 2016 and Amazon introduced Transparent Ad Marketplace (“TAM”) at the end of 2016. In these alternatives, the auction among SSPs takes place in a remote server controlled by a third party (the provider of the server-side header-bidding solution) instead of in the user’s browser. This helps to improve site-load speed. On the other hand, this solution leads to less revenue for publishers and reduces the availability of data to advertisers and publishers.[55]

Over the past decade, the price of digital advertising has fallen steadily, while output has risen. U.S. digital-ad spending grew from $26 billion in 2010 to $189 billion in 2021, an average annual increase of 20%.[56] Over the same period, the Producer Price Index for Internet-advertising sales declined by an annual average of 4%.[57] The rise in spending in the face of falling prices indicates that the number of ads bought and sold increased by approximately 25% annually. The combination of increasing quantity, decreasing cost, and increasing total revenues are consistent with a growing and increasingly competitive market, rather than one of rising concentration and reduced competition.

D.   Digital advertising is a multisided market

The digital advertising market can be thought of as a complex multi-step and multisided market that involves three key parties—advertisers, publishers, and intermediaries—and is aimed at a fourth: consumers. In contrast, critics of the current structure of and conduct in the digital advertising industry have characterized it as a “straightforward and traditional” market in which publishers supply an inventory of ad space and advertisers are buyers of the ad space.[58] In this simplistic account of the market, for a given supply of inventory, publishers would seek to maximize the price received per ad, while advertisers would seek to minimize the price paid per ad. Targeting of ads would be based on the demographics of a publisher’s readership or the content of the publication, rather than the individual characteristics of each reader. In general, this is how the market initially operated before the introduction of clickable ads. But even this simple formulation is quite complex. Advertisers expect to maximize the return on their investment in advertising. Even at a low price, advertising expenditures would be wasted if that investment were not converted to increased sales of the advertiser’s product or service.

The invention of clickable ads with which users could interact changed the objective function of digital advertising. Publisher revenues and advertising costs became linked to individual consumers acting on an ad by, for example, clicking on it. Rather than paying or receiving a price-per-ad based on the size of a publication’s user base, advertising expenditures became a function of a price-per-click (or other action) and the number of clicks. This meant that the rewards for relevance—as well as the complexities of determining relevance—were greater because some viewers might be persuaded to act there and then.

In this multisided market, ad intermediaries must balance the interests of at least three constituencies: (1) advertisers creating ads and placing them; (2) publishers defining inventory and displaying ads; and (3) users consuming published content who view and act on ads. Intermediaries in these markets often benefit from network effects, through which the value of the platform to each user depends in part on the number and quality of other users on the platform.[59]

The quality and relevance of users is assessed by collecting information on the users as they browse the web. This information can include which ads they have viewed and clicked in the past, their geographical location, as well as their demographics, financial situation, and topics of interest. Broadly speaking, a larger network with diverse users provides more information and is better able to target ads to relevant users, benefiting advertisers, publishers, and consumers.

Network effects are not always positive, however, nor are they always captured by the platform that facilitates them.[60] While access to consumer data can help to improve the quality of the ads displayed—and increase the value of those ads to advertisers and publishers—claims that such access provides increasing returns to scale are not borne out by the burgeoning empirical literature on the topic. Summarizing these empirical findings, economist Catherine Tucker concludes that “empirically there is little evidence of economies of scale and scope in digital data in the instances where one would expect to find them.”[61]

Intermediaries in multisided markets often face difficult optimization problems caused by the interrelated demands of participants on different sides of the market, each group of whom benefits from the existence and size of the other, but whose interests conflict across many margins.[62] This highlights the key distinction between “straightforward and traditional” markets and multisided markets.

Ad tech intermediaries that are vertically integrated into some or all components of the ad tech stack use prices charged to each side of the market to optimize overall use of the platform. As a result, pricing in these markets operates differently than pricing in traditional markets. Pricing on one side of the platform is often used to subsidize participation on another side of the market, increasing the value to all sides combined. Consequently, pricing (or other terms of exchange) may appear to one side of the market to diverge from the competitive level when viewed for that side alone. While one side of the market may pay higher fees, this cost can be offset by the benefits from increased participation on the other side of the market. Thus, using subsidies to increase participation on another side of the market creates valuable network benefits for the side of the market facing the higher fees.

For example, among the criticisms of digital advertising business practices is the use of “second-price auctions” rather than “first-price auctions.”[63] First-price auctions are those most familiar to people: multiple bidders offer prices, and the highest bidder wins the auction and pays an amount equal to her winning bid. In a second-price auction, the highest bidder wins the auction but pays an amount based on the next-highest bid. In markets with many bidders possessing the same information, first-price auctions and second-price auctions would be expected to produce the same amount of revenue under the well-known auction-theory concept of revenue equivalence.[64]

The choice of auction approach reflects the tensions between different sides of the market in a multisided market. On the one hand, under certain circumstances, a first-price auction tends to increase the prices received by sellers (here, publishers), but exposes buyers (here, advertisers) to an increased risk of overpayment.[65] On the other hand, under certain conditions, a second-price auction reduces risks to advertisers, but also reduces the prices received by publishers.[66] It would be expected that an ad tech intermediary would balance these competing interests to maximize total revenues flowing through the ad tech stack, to maximize its profitability. In such a multisided market, it would be a mistake to focus only on one side of the market and ignore the effects that decisions such as this have on the other participants.

The extent to which ad tech intermediaries—in particular, vertically integrated services like Google’s—act to optimize the overall value of the platform is critical to understanding how these markets work. It also highlights how misleading it can be to assume that these processes can be analyzed as “straightforward and traditional” markets.

II. Antitrust Primer: Effective Competition Is not an Antitrust Offense

A flawed premise underlies much of the Texas Complaint, the Omidyar Network’s Roadmap report, and the CTDAA legislation. Fundamentally, most of the charges that each of these level against Google and Facebook’s ad tech businesses derive from an assertion that conduct engaged in by dominant incumbent firms that makes competition more difficult for competitors is anticompetitive—even if the conduct confers benefits on users. This often amounts to a claim that the largest firms are acting anticompetitively by innovating and developing their business processes and products in ways that create benefits for one or more digital advertising constituents and for the advertising ecosystem more generally. These include creating new and innovative products, lowering prices, reducing costs through vertical integration, and enhancing interoperability between existing products, among other things.

This approach entails an argument—made explicit in the Texas Complaint and the Omidyar Roadmap—that Google harms competition by creating obstacles for rivals without offsetting “incremental efficiencies.”67F[67] According to the report, this means that, even if Google’s practices produce benefits for such constituents as advertisers, publishers, or consumers, they could possibly be reimagined to create even more competition or achieve the same benefits in ways that better prop up rivals. According to the Roadmap, the practices should therefore be condemned as anticompetitive.68F[68]

But that is not how the law—or the economics—works. Such an approach converts manifestly beneficial aspects of Google’s ad tech business into anticompetitive defects, essentially arguing that successful competition creates barriers to entry that merit correction through antitrust enforcement. The CTDAA takes this argument a step further by imposing “best interests,” “best execution,” and “transparency” obligations on large firms and mandating divesture of parts of the largest firms to facilitate more entry by competitors. This approach turns U.S. antitrust law (and basic economics) on its head. As some of the most famous words of U.S. antitrust jurisprudence have it:

A market may, for example, be so limited that it is impossible to produce at all and meet the cost of production except by a plant large enough to supply the whole demand. Or there may be changes in taste or in cost which drive out all but one purveyor. A single producer may be the survivor out of a group of active competitors, merely by virtue of his superior skill, foresight and industry. In such cases a strong argument can be made that, although, the result may expose the public to the evils of monopoly, the Act does not mean to condemn the resultant of those very forces which it is its prime object to foster: finis opus coronat. The successful competitor, having been urged to compete, must not be turned upon when he wins.69F[69]

U.S. antitrust law is intended to foster innovation that creates benefits for consumers, including innovation by incumbents. The law does not proscribe efficiency-enhancing unilateral conduct on the grounds that it might also inconvenience competitors, or that there is some other arrangement that could be “even more” competitive. Under U.S. antitrust law, firms are “under no duty to help [competitors] survive or expand.”70F[70]

To be sure, the arguments are couched in terms of anticompetitive effect rather than being described merely as commercial disagreements over the distribution of profits. But these effects are simply inferred, based on assumptions that Google and Facebook’s vertically integrated business models entail an inherent ability and incentive to harm rivals. For example, Google is alleged to be able to surreptitiously derive benefits from display advertisers by “leveraging” its search-advertising capabilities,71F[71] or by “withholding YouTube inventory,”72F[72] rather than altruistically opening it up to rival ad networks, or by using its access to data to improve its products without sharing that data with competitors.

All these charges may be true, but none is inherently anticompetitive. Under U.S. law, companies are not obligated to deal with rivals and certainly are not obligated to do so on rivals’ preferred terms.73F[73] In the Texas Complaint, for example, the court, citing Charych v. Siriusware, noted, “[D]efendants were under no obligation to develop an interface that was compatible with plaintiffs’ product.”74F[74] As long ago as 1919, the U.S. Supreme Court held that “[i]n the absence of any purpose to create or maintain a monopoly, the [Sherman Act] does not restrict the long recognized right of [a] trader or manufacturer engaged in an entirely private business, freely to exercise his own independent discretion as to parties with whom he will deal.”75F[75] More recently (in 2004) the Court held:

Firms may acquire monopoly power by establishing an infrastructure that renders them uniquely suited to serve their customers. Compelling such firms to share the source of their advantage is in some tension with the underlying purpose of antitrust law, since it may lessen the incentive for the monopolist, the rival, or both to invest in those economically beneficial facilities.76F[76]

Moreover, U.S. antitrust law does not second guess unilateral conduct simply because it may hinder rivals. Any such conduct would first have to be shown to be anticompetitive—that is, to harm consumers or competition, not merely certain competitors.77F[77] In two-sided markets, this means finding not simply that some firms on one side of the market were harmed, but rather that the combined net effect of challenged conduct across all sides of the market was harmful.78F[78]

In the platform context, understanding whether there is harm to competition at all requires an assessment of the effects of conduct on all sides of the platform. “[N]o economic basis exists for establishing a presumption that ‘harm’ on one side of a two-sided platform is sufficient to demonstrate that market output has been restricted, or that consumer welfare has otherwise been harmed.” In fact, “[s]eparating the two markets allows legitimate competitive activities in the market for general purposes to be penalized no matter how output-enhancing such activities may be.”79F[79]

The Texas Complaint, however, is built on the alleged harm of reduced revenue to publishers, without considering the corresponding benefit of lower prices to advertisers, or the net effect on consumers.80F[80]

Beyond that, antitrust law does not condemn conduct on the basis that an enforcer (or a court) is able to identify or hypothesize alternative conduct that might provide similar benefits at lower cost. In alleging that there are ostensibly “better” ways that Google could have pursued its product design, pricing, and terms of dealing, both the Texas Complaint and Omidyar Roadmap do just that—assert that, had the firm only selected a different path, an alternative could have produced even more benefits or an even more competitive structure. This line of thinking seems to be one motivation for the CTDAA’s remedies.

The reason that the possibility of “better” theoretical arrangements cannot serve as the basis for antitrust intervention is that there are limits to what can be achieved through intervention, not least because of limitations on legislators’ and enforcers’ knowledge about the competitive dynamics of the markets they seek to regulate.81F[81] A practice’s departure from a theoretical competitive benchmark may be inextricably linked to the social benefits it generates. When this is the case, enforcement that requires the practice or product to change in order to adhere to a theoretical standard may end up undermining the benefits of the practice in the first place. That is particularly true in the context of the sort of “vertical foreclosure” arguments leveled against Google in the advertising space, in which it is alleged that the combination of different levels of the ad-supply chain by Google limits the ability of competitors to enter and compete effectively.82F[82] It is surely conceivable that the product improvements conferred by the combination of different functions into a single platform—e.g., greater efficiency, realization of network effects, more effective targeting—could be replicated by a different means that might also facilitate “even more competition.” But such an approach is fraught with the risk of serious and costly error.83F[83]

The alleged cure of tinkering with benefit-producing unilateral conduct by applying an “even more competition” benchmark is worse than the supposed disease. The adjudicator is likely to misapply such a benchmark, deterring the very conduct the law seeks to promote. As then-Judge Stephen Breyer explained in the context of above-cost low pricing (another “defect” that both the Texas Complaint and the Roadmap claim constitutes anticompetitive conduct by Google84F[84]), “the consequence of a mistake” is “to penalize a procompetitive price cut,” conduct that, from an antitrust perspective, is “the most desirable activity.”85F[85] That commentators or enforcers may be able to imagine alternative, theoretically more desirable, conduct is beside the point.

Similarly, subjecting the kinds of product-design decisions at issue in the Google case to refined balancing of benefits and harms would deter innovation. “To weigh the benefits of an improved product design against the resulting injuries to competitors is not just unwise, it is unadministrable. There are no criteria that courts can use to calculate the ‘right’ amount of innovation, which would maximize social gains and minimize competitive injury.”86F[86] Put simply, “no one can determine with any reasonable assurance whether one product is ‘superior’ to another.”87F[87]

For these reasons, a “product improvement by itself does not violate Section 2, even if it is performed by a monopolist and harms competitors as a result.”88F[88] “Any other conclusion would unjustifiably deter the development and introduction of those new technologies so essential to the continued progress of our economy.”89F[89] A benefit-creating product design, even if it hinders rivals, is “necessarily tolerated by the antitrust laws.”90F[90]

Nor does U.S. law condemn a firm’s decision not to share a product improvement with rivals on terms rivals might prefer, even when such sharing might lead to greater competition in the short term. “Compelling” innovators “to share the source of their advantage” with rivals, among other evils, “may lessen the incentive for the monopolist, or rival, or both” to invest in innovation.91F[91] Except in extremely limited circumstances, firms can decide the terms on which they offer their products and services.

Directly rejecting the Roadmap’s suggestion—and the CTDAA’s mandate—of compelling dealings on terms that might produce greater competition, the Supreme Court has decreed that the “Sherman Act . . . does not give judges carte blanche to insist that a monopolist alter its way of doing business whenever some other approach might yield greater competition.”92F[92] Firms are not obliged to go into new lines of business or abandon existing lines to throw lifelines to rivals.93F[93]

The law similarly encourages vertical integration, because it tends to foster innovation-enhancing synergies and lower prices by eliminating double marginalization.94F[94] As the Roadmap elsewhere admits, it is “not in itself uncommon” to see vertical integration result in “fewer and fewer companies,” even “in competitive markets.”95F[95] Thus, vertical integration by internal expansion—even by a monopolist—is presumptively lawful.96F[96] The Roadmap and the CTDAA, however, simply disregard this, instead presumptively condemning vertical integration that hinders rivals by creating efficiencies.97F[97] Again, this is simply not a defensible interpretation of U.S. antitrust law, nor should it be.

III. Allegations Against Google in Digital Display Advertising

Critics of the digital advertising industry—and Google’s role in it—have leveled numerous allegations. These include claims that Google “leverages” its ownership of YouTube to obtain and exert market power in the buying and selling of other digital-display ads. Some claim that Google anticompetitively uses cross-subsidies, charging supercompetitive prices at one end of the ad tech stack to subsidize supra-competitive prices at another end of the stack. It is also alleged that Google has superior information about consumers that it will not provide to competitors, giving Google an anticompetitive advantage. It is claimed that steep entry barriers—some allegedly erected by Google—inhibit entry and allow the company to achieve a supercompetitive “take rate” from its intermediation services. While the lawsuits may provide additional information and data to support these claims, we argue that, with the limited public information available to us, it is not clear that any of them constitute anticompetitive conduct.

A.   ‘Leveraging’ market power in video streaming into the digital open-display market

The Omidyar Roadmap argues that Google, by virtue of its vertical integration throughout the intermediary stack and into the supply side (as the owner of YouTube), has the incentive and ability to derive unwarranted benefits from its display advertising business. It alleges, for example, that, by offering a single interface for placing both search and display ads, “Google leverages its monopoly position in search to coerce advertisers into using Google’s display products.”98F[98] In support it cites the CMA as saying:

Google may also be able to leverage its market power in search into the open display market. Smaller advertisers often choose to single-home to minimize transaction costs. Advertisers that wish to single-home have a strong incentive to use Google Ads as they can use it to access Google search advertising and YouTube inventory as well as the open display market.99F^[99]

An earlier version of the Texas Complaint echoed these claims:

Google’s practice of withholding YouTube video inventory from rival ad buying tools… effectively locks single-homing small advertisers into Google’s ad buying tool. In addition, other providers of ad buying tools indicate that it does not make economic sense to try to compete with Google Ads for small advertisers, because they cannot achieve sufficient scale with smaller advertisers who want to buy display, YouTube, and even search ads, through just one tool.100F[100]

And, similarly, the Roadmap also argues that most sources of demand for Google Ads purchase ad space through AdX because Google “designed its exchange in such a way that it operates more efficiently with requests from Google’s own ad server than it does when requests come from rival ad servers.”101F^[101]

All these assertions describe efficiency-enhancing behavior as anticompetitive. The report does not allege that Google preferences its own ad exchange in ways that harm advertisers; rather, the company’s products simply work better together (which is not unusual when different software products must interact) and it is thus in advertisers’ best interests for Google to act this way.

U.S. law, rightly, does not consider efficiencies obtained from vertical integration in this way to be anticompetitive. Nor do efficiencies that rivals cannot beat qualify as “barriers to entry.” The alternative—requiring Google to refrain from using the cheapest and/or fastest option available, because doing so makes its product better than all competitors—would mean reduced innovation, higher overall costs, and no benefit to either advertisers or publishers.

Later, the Roadmap makes another similar allegation: that Google “leverages” its ownership of YouTube, and the fact that only Google’s DSP can place ads on YouTube, to give itself an anticompetitive advantage in open-display advertising because rival DSPs are inherently limited by being unable to place ads on YouTube. An earlier version of the Texas Complaint echoed this claim.102F[102]

The Roadmap characterizes this conduct as “a contractual way to deny interoperability,”103F[103] but there is no contractual restraint here. How Google distributes YouTube’s ad inventory is a unilateral distribution decision permitted under U.S. law. And Google’s policy is not unusual in any way: many other websites that carry video advertising—including Hulu, Instagram, and Twitter—self-distribute their own inventory and do not make it available for resale by third parties.104F[104] Google does not have a duty to maximize its competitors’ profits by allowing them to resell YouTube inventory.

Access to YouTube is also not essential to a DSP’s success. Before Google stopped third-party platforms from buying YouTube ad inventory, it reported that only a “small amount” of buying was being done through Google’s AdX, which allowed third-party platforms to bid. At the time, AdExchanger reported that “[b]y ’small amount,’ that reportedly means 5%.”105F[105] A competing DSP, TubeMogul, said that this decision was an “unfortunate development” but “immaterial, since less than 5% of total ad spend through our software in Q2 was directed to YouTube.”106F[106]

This is consistent with the fact that there are several successful DSP competitors that compete with Google, despite not having access to YouTube’s ad inventory. The Trade Desk went public for $1 billion in 2016, processed more than $6.2 billion in transactions in 2021, and had a market cap of more than $25 billion in the first week of August 2022. 107F[107] Other DSPs, like Amazon’s and Xandr (formerly AppNexus), both continue to compete with Google vigorously without access to YouTube inventory, as the Omidyar Roadmap admits in the case of AppNexus.108F[108]

The Roadmap further -alleges that Google’s owned-and-operated properties—including Search, YouTube, Shopping, Flights, and News—confer an anticompetitive advantage because “Google pays no ‘traffic acquisition costs’” for the advertising space on its own sites: “When Google places ads on YouTube, just as when it places ads on its own search results pages, Google pays no ‘traffic acquisition costs’ because it needn’t pay any publisher for access to the ‘eyeballs’ that will see or interact with the ads it helps place.”109F^[109]

Google’s parent Alphabet reported that the company’s traffic-acquisition costs were approximately 20% of its revenues in 2021.110F^[110] Over the past few years, 40-50% of Alphabet’s expenditures have been on “cost of revenues,”111F[111] and of these, roughly half have involved traffic-acquisition costs.112F[112] Alphabet defines traffic-acquisition costs as (a) “the amounts paid to our distribution partners who make available our search access points and services” and (b) amounts paid for ads displayed on Google Network Members properties. It identifies “distribution partners” as browser providers, mobile carriers, original equipment manufacturers, and software developers.

Contrary to the Roadmap’s insinuations, there is nothing to suggest that these expenditures become less burdensome as a company increases in scale. Indeed, the opposite may be true, if it is more costly to gain access to marginal users than inframarginal ones, consistent with Google’s traffic-acquisition costs increasing over the years as it has grown.113F[113] While Google does not have to pay itself for the use of its own display inventory, there is clearly an opportunity cost to displaying its own inventory rather that of another firm. The claim that the company faces no traffic-acquisition costs for these properties is inaccurate.

The Roadmap’s focus on traffic-acquisition costs also overlooks content-acquisition costs—the payments to content providers from whom Google licenses video and other content for distribution on ad-driven and subscription services such as YouTube and Google Play. While Google does not pay a publisher for access to “eyeballs” on its owned-and-operated properties, it pays substantial and increasing amounts for content on those properties that attract the “eyeballs.”114F^[114] Alphabet CFO Ruth Porat indicates, for example, that YouTube pays content creators “a majority of our revenue.”115F[115] Leaving this expense out of the calculation is another example of the over-simplification that characterizes many of the claims that Google’s ad tech business is a simple (and simply anticompetitive) business.

B.   Excess pricing

Where Google’s critics diverge most significantly from the spirit of U.S. antitrust law is in their overriding concern for how advertising revenues are distributed among the recipients of advertisers’ payments: intermediaries (Google) and publishers. The Texas Complaint alleges that Google has a higher “take rate” than competing exchanges,116F[116] is able to increase its take rates without losing market share,117F[117] and “manipulates auctions to increase its take rate.”118F[118] This follows the Roadmap’s speculation—based on the CMA Interim Report—that Google may take a larger cut of advertising spending than its competitors.119F^[119] And these allegations echo claims made in another report that Google introduces “hidden” fees that increase the overall cut it takes from ad auctions.120F[120]

First, it should be noted that the basis for these claims in the Roadmap are drawn from the CMA investigation’s interim report, published in December 2019. In the final report, after further investigation, the CMA abandoned this claim. The final report describes the CMA’s analysis of all the billions of Google Ad Manager open auctions related to U.K. web traffic during the period between March 8 and March 14, 2020. This, according to the CMA, allowed it to observe any possible “hidden” fees, as well. The CMA concludes:

Our analysis found that, in transactions where both Google Ads and Ad Manager (AdX) are used, Google’s overall take rate is approximately 30% of advertisers’ spend. This is broadly in line with (or slightly lower than) our aggregate market-wide fee estimate outlined above. We also calculated the margin between the winning bid and the second highest bid in AdX for Google and non-Google DSPs…. We found that Google’s average winning margin was similar to that of non-Google DSPs. Overall, this evidence does not indicate that Google is currently extracting significant hidden fees. As noted below, however, it retains the ability and incentive to do so.121F^[121]

This is a crucial finding that severely undermines the allegations that Google extracts excessive or “hidden” fees. It also undermines the claim that there are “missing funds accruing to Google.” While these conclusions do not eliminate the possibility that the industrywide price could itself be above competitive levels (and it remains to be seen whether the plaintiffs states in the Texas case will produce different evidence), they do mean that the best evidence currently available calls into question the charge that Google exploits a lack of interoperability by prioritizing its own products or that it engages in opaque pricing to conceal hidden charges of which its customers are unaware.

More fundamentally, absent evidence of Google deceiving advertisers and publishers to extract above-competitive margins, claims that its prices are “too high” or its revenue sharing “too low” are at odds with established antitrust law. U.S. antitrust law does not attempt to derive “proper” prices and impose these obligations on companies to ensure a “fair” outcome. Absent anticompetitive defects in the process, even monopolists are free to charge monopoly prices. The alternative would be for some agency—a court or a regulator—to regulate pricing and second guess every business decision made by dominant firms.

C.   Cross-subsidies

At the same time, the Roadmap alleges that Google can “charge low prices at one end of the stack, to drive out competitors, while charging high prices at the other to counterbalance any losses.”122F[122] But even if true, this would not be anticompetitive. It is a widely understood feature of platforms that they can shift prices from one side of a multisided market to another to maximize the platform’s total value. For example, a marketplace may make sellers bear the burden of fraud or mis-selling to give assurance to customers, and grow the consumer side of the platform market, just as a ridesharing app may discount rides to attract customers to build a large enough base to induce drivers onto the app.

This is a normal part of platform economics, which has long recognized that offering one side a low, zero, or negative price can be efficient and procompetitive.123F[123] As the U.S. Supreme Court held in Ohio v. American Express, an integrated competitive-effects analysis should look at the overall effect on output, not the effect on one side of the market; the relevant market must include both sides of the platform or the market would not exist at all.124F[124] There is no reason to think that this kind of behavior would generally be classed as “predatory pricing” in the absence of other behavior, such as raising prices after driving out competitors.125F[125]

But neither the Texas Complaint nor the Roadmap allege that Google’s prices were predatory. On the contrary, their sole claim in this respect is that, after being acquired by Google, DoubleClick lowered its prices (by a factor of ten, according to the Roadmap126F[126]), which it then maintained at these lower levels. This price reduction is facially procompetitive, however. It is unusual, to say the least, to describe a price reduction, with no subsequent price rises, as anticompetitive. If less-efficient competitors were unable to compete with these lower prices, that is competition in action. The law does not preclude nonpredatory low prices, nor even predatory prices without recoupment.127F[127] Sustained price reductions are one of the primary goals of antitrust.

Moreover, the source of the Roadmap’s claim that these price reductions were done “to drive out competitors” was, notably, a company that was not actually driven out of business by these price reductions. The source was an ad server, Smart, which claimed that Google’s price reductions “made the provision of publisher ad server difficult to sustain as a standalone business. This was the main reason why Smart felt the need to expand into the provision of SSP services.”128F[128] A competitor of Google’s responding to price reductions by broadening its own offerings is, again, procompetitive, not anticompetitive.

D.   Data gathering and integration

The Texas Complaint and the Roadmap describe several pro-privacy measures Google has adopted or plans to adopt as being detrimental to its competitors, including the decision to disable third-party cookies (which allow digital advertising companies to track users across the web to serve them relevant targeted ads) on the Chrome browser.129F[129] The Complaint argues that this shift benefits Google to the detriment of other ad tech companies, because (it says) Google, but not its competitors, has other data sources it can use to target ads at users.130F[130] In the same vein, the Roadmap points to Google’s decision not to share with advertisers raw data that it compiles about users.131F[131]

The Complaint ignores regulatory causes of these changes altogether, and the Roadmap dismisses the suggestion that they may be driven by the European General Data Protection Regulation (GDPR), on the basis that “data sharing here in the U.S., where we have no privacy regulatory scheme akin to that which is in place in Europe” has also been curtailed.132F[132] Both forget data-privacy laws in U.S. states, such as the California Consumer Privacy Act (CCPA). And, even if that weren’t the case, the claim that GDPR would have no effect outside Europe ignores that companies may find it easier to comply with such laws by changing their practices globally, rather than on a country-by-country or state-by-state basis.133F[133] Many companies have done this: Microsoft, for example, announced in November 2019 that it would “honor California’s new privacy rights throughout the United States.”134F[134]

Both also ignore the possibility that these provisions may be a response to demand from users of Google Chrome. Google may have good reasons to maintain a reputation for protecting user privacy, particularly because of the wide range of services it provides where user privacy is often of paramount importance to many users: Search, Maps, Gmail, YouTube, and Chrome itself.

Apple and Mozilla, neither of which has a significant online display advertising arm (and thus, have no incentive to block cookies simply to disadvantage display advertisers, as the Complaint alleges Google has done) have taken similar steps to increase user privacy. These are direct competitors of Google Chrome’s, and when Apple made blocking third-party cookies the default in its Safari browser, it was reported by one major outlet as “beating Google by two years to the privacy feature.”135F[135] Indeed, one of the reasons that Google delayed its disabling of third-party cookies was reportedly to implement technologies to “make it easier for advertisers to target certain demographics without laser-sighting down to specific people, ensure that the infrastructure many sites use for logins don’t break, and help provide some level of anonymous tracking so advertisers can know if their ads actually converted into sales.”136F[136]

That Chrome’s competitors, neither of which has an incentive to hurt ad tech companies, have taken the same steps that the Complaint alleges Google is taking for anticompetitive reasons should be compelling evidence that Google, too, is responding to user demand and/or regulation. Under U.S. law, the fact that these are legitimate moves and benefit users interested in privacy—and, indeed, may be a response to competition in the browser market—undermines claims that Google has failed to maximize competition along other dimensions.

The Roadmap also presents a hypothetical circumstance that amounts to an allegation that Google “captures” data from ads served to publishers to “expropriate” publishers’ investments in content that attracts a particular audience:

Some publishers have invested in content that attracts and retains a specific type of consumer, for example, readers of the Wall Street Journal or Golf Digest; this in turn allows them to support their business by selling valuable ads to advertisers looking for exposure to those audiences. Google has two ways to expropriate that value. First, rather than serve an ad on the Wall Street Journal at a high price, it can track the user who visited the Wall Street Journal and wait until she visits a site that sells space at low prices, for example, a local recipe blogger. Google can then sell Wall Street Journal users to advertisers in a way that does not benefit the Wall Street Journal at all and costs advertisers much less. A second strategy used by Google is to take the data describing these differentiated audiences and use it to create an imitation portfolio of consumers that mimic the characteristics of the publisher’s audience. For example, Google could create an audience of consumers similar to the people who read Golf Digest. Then Google sells access to this group of consumers when they visit inexpensive websites. Advertisers are happy to buy these ads because the consumers likely belong to the specialized audience of interest but are available at a much lower price. In these ways the unique audience assembled by the publisher is copied and expropriated.137F[137]

It should be noted that the Roadmap does not conclude that Google engages in these practices, but merely describes strategies Google “can” undertake to “expropriate” publishers’ investments. The Roadmap concedes that advertisers would be “happy” under such hypotheticals, because they are buying effective ads at a “much lower price.” In the Roadmap’s example, the hypothetical recipe blogger is “happy” that it earned revenues from selling an impression and the advertiser is “happy” that it paid a lower price than it would have had the impression been sold to the Wall Street Journal. The Wall Street Journal may not be so “happy” that it did not serve that particular ad, but that display space did not sit empty; the hypothetical lost ad was replaced by another impression that was served. And it is neither Google’s nor antitrust law’s job to make specific publishers better off—nor to make publishers better off at the expense of advertisers—but to ensure that the market as a whole is competitive and acting in consumers’ interests.

These hypotheticals again highlight the tensions discussed above between the different sides of a multisided market. Actions that make advertisers “happy” may come at the expense of publishers’ advertising revenues and actions that increase publishers’ revenues may increase costs to advertisers. One of the goals of a multisided market intermediary such as Google is to balance these competing interests to maximize total revenues flowing through the ad tech stack.

The Roadmap concludes that, through its “entire family of products,” Google collects and analyzes substantial amounts of information about its users. It uses this information to maximize the “effectiveness,” “precision,” and “value” of the ads it intermediates.

First, Google offers an entire family of products—everything from Gmail and Google Maps to the Google Calendar, Google Chrome, Android mobile operating system and the search engine—that gather valuable personal data about its users. Second, the products across the ad stack further collect data on consumer activities that the company then integrates to maximize the effectiveness and precision of ad targeting and attribution and thereby the value of the ads.138F[138]

Rather than “expropriating” publishers’ data, it would be reasonable to conclude that Google is adding value to the data provided by publishers, advertisers, and consumers to better target ads. For example, the Wall Street Journal may not know that a consumer recently did a Google Search for “running shoes.” By adding valuable information from Search, the consumer might be served a relevant running shoe ad on the Wall Street Journal’s site. This benefits the publisher who is paid for serving a valuable impression, the advertiser who sells a pair of shoes, and the consumer who obtains useful information and purchases the product she was seeking.

E.   Accelerated Mobile Pages (AMP) and header bidding

The Texas Complaint, like the Roadmap, alleges that Google designed Accelerated Mobile Pages (AMP) “[t]o respond to the threat of header bidding… [by making it] essentially incompatible with JavaScript and header bidding. Google then used its power in the search market to effectively force publishers into using AMP.”139F[139] But this gets several key facts wrong. First, AMP is an open-source industry collaboration project and Google cannot unilaterally impose a design standard on it. 140F[140] Second, a version of header bidding can work with AMP.141F[141] Third, it is mistaken to assert that “non-AMP-formatted results often do not even show up on the first page of results, regardless of their relevance.”142F[142] AMP has been a prerequisite only for inclusion in the top news story carousel, while other listings are ranked by relevance and speed.

Importantly, the argument ignores the main benefit of AMP to publishers and users: faster load times for mobile users who may be on slow connections. One of header bidding’s chief downsides is that it increases page-load latency. It is obvious why an HTML framework built to maximize load times would not be compatible with header bidding. Because AMP confers undisputed benefits on users and publishers, Google and the other companies involved in the AMP project have no obligation to re-engineer AMP to be compatible with header bidding. Any conclusion otherwise would involve a court deciding that users should be forced to use a slower Internet so that websites can use header bidding.

F.   Alleged barriers to entry in the open digital display ad market

Claims about Google’s alleged market power in display advertising rest on assumptions that the company enjoys the benefits of significant barriers to entry throughout the ad tech stack, thus enabling it to extract supercompetitive rents without fear of competition: “With these barriers in place,” it is claimed, “entry seems nearly futile.”143F[143]

A key element in establishing a company’s durable market power—and thus, its ability to impose anticompetitive costs on its users—is the presence of entry barriers. Even a market with only a single company—a true monopoly—cannot act like a monopoly if entry into its market is easy; if it did profitably raise prices, new competitors would enter the market and undercut it.144F[144]

As the Roadmap concedes, “[m]arket power is not permanent, of course. It can be undercut by, among other things, new entrants that offer better quality or lower prices.”145F[145] This notion of “contestability” is a fundamental part of assessing the competitiveness of markets under U.S. antitrust law.146F[146] In the absence of barriers to entry, it is well-established that assumptions of future competitive harm from ongoing conduct cannot be sustained.147F[147] Thus, the Roadmap bases much of its brief against Google on the presence of barriers to entry, “which heighten[] the prospect that Google can engage in conduct that harms competition without restraint from new entrants or potential new entrants.”148F[148] On the strength of these asserted barriers, the Roadmap’s authors interpret ambiguous conduct as anticompetitive.

According to the Roadmap, “the CMA’s findings reveal a number of significant barriers to entry into the digital advertising market.”149F[149] But most of its assertions in this regard are flawed, either because the CMA did not, in fact, make “findings” in the ways it suggests, or else because it reaches incorrect conclusions that certain conduct constitutes a barrier.150F[150] The Texas Complaint’s assertions of similar barriers to entry are likewise problematic.151F[151]

1.     Consumer location information

Although the Texas Complaint does not discuss it as an explicit barrier to entry,152F[152] one of the Omidyar Roadmap’s key assertions about barriers to entry relates to Google’s access to user-location data. It asserts that:

The CMA concluded that Google has nearly insurmountable advantages in access to location data, due to the location information it receives from the Android operating system, Google search, and other applications…. An entrant into the ad tech stack requires information about the consumer to target an ad effectively. Because Google accounts for nearly the entirety of the mobile search sector in the UK—97%—and controls many of the known sources of location data, such an entrant faces a large barrier to entry.153F[153]

But the CMA does not, in fact, “conclude[] that Google has nearly insurmountable advantages in access to location data,” either in the CMA Interim Report to which the Roadmap refers, nor in the CMA Final Report. The CMA never makes any claim of “insurmountable advantage.” Indeed, it does not use the word “insurmountable” at all, except to note that “rival platforms did not suggest that accessing consumer data was an insurmountable barrier to entry.”154F[154] Rather, to support this claim, the Roadmap cites to a portion of the CMA Interim Report recounting a suggestion made by Microsoft regarding the “critical” value of location data in providing relevant advertising.155F[155]

Moreover, that portion of the CMA Interim Report, as well as the suggestion made by Microsoft, is about search advertising, not display advertising. While the CMA does not characterize this data in the way the Roadmap claims, it does allege that “Google has exclusive access to a large amount of user data that can be used for targeted advertising and for measuring advertising outcomes, collected through its consumer-facing services. Data collected on its search platform is particularly valuable for targeting purposes in open display as it reveals users’ purchasing intent.”156F[156]

While location data may also be valuable for display advertising in a comparable way, it is not clear that the GPS-level data that is so valuable in providing mobile-search ad listings is particularly useful for display advertising, which may be just as well-targeted by less granular city- or county-level location data.

Consider the Roadmap’s illustrative example:

A digital ad for a brick-and-mortar running store in Des Moines is of little use to a runner looking to test out new shoes in Omaha, and, if shown to the Omaha runner, is unlikely to prompt a click, much less a purchase, from the Des Moines store.157F[157]

This is certainly correct. But GPS or even cell-tower location data is not necessary to determine in which city a user is located. Publicly available databases of IP address locations can provide this information, and they are readily and often freely available to all competitors. It is difficult to imagine that display advertising uses location data at any greater level of granularity except in unusual circumstances; it simply would not be particularly useful or effective.158F[158]

Furthermore, to the extent that location data (like other consumer data) may be useful for display advertising, the most significant issue affecting its availability to advertisers is not Google’s presence in the ad tech stack; it is privacy regulations that limit the collection, use, and sharing of such data.159F[159] These privacy regulations, such as the GDPR, limit the ability of digital firms to sell or otherwise pass user data to third-party advertisers. What seems like unequal treatment, in this regard, is really a case of privacy regulation in action.

These laws may have the indirect effect of favoring larger digital conglomerates that can collect user data through one service and use it to target ads in another. In this sense, it could be true that Google has informational advantages over rivals, though in a critically different way than that alleged by the Roadmap. But it can hardly be considered anticompetitive if the source of such advantage is legal constraints on information sharing. Indeed, an empirical study by economists Avi Goldfarb and Catherine Tucker found that (pre-GDPR) privacy regulation in the EU “restricted advertisers’ ability to collect data on Web users in order to target ad campaigns. We ?nd that, on average, display advertising became far less effective at changing stated purchase intent after the EU laws were enacted, relative to display advertising in other countries.”160F[160] Along similar lines, Nils Wernerfelt and his co-authors show that access to data from different sources significantly improves ad targeting.161F[161] In turn, this may give a competitive advantage to firms that operate several successful web services and applications.

As Israeli competition law scholars Michal Gal and Oshrit Aviv found with respect to the GDPR, privacy regulations can function as a barrier to entry in several further ways.162F[162] These include creating new economies of scale associated with regulatory compliance, increased litigation risk, and uncertainty around interpretation of the rules. Because they serve to make reputation more central, they also can lead users to become more likely to entrust their data to incumbents but not to unknown, new entrants.163F[163]

2.     Attribution measurement

“Attribution” refers to the method by which advertisers can see which ad led a user to an action, such as visiting a website or making a purchase.164F[164] The Roadmap alleges that Google can design attribution to mislead advertisers by, for example, favoring search ads over display ads.165F[165] This would lead to more of the advertiser’s money going to Google instead of (in part) to a publisher, and (assuming, as the Roadmap implies, that this makes ad campaigns less effective166F[166]) can harm advertisers by misleading them into choosing a less-effective advertising channel.

The Roadmap provides no evidence this is taking place. What it describes is more a complaint about the nature of search advertising in general: that companies will sometimes end up paying for ads in lieu of identical organic search results for their pages. That is an argument to be had elsewhere, but there are clear reasons why it may be in an advertiser’s interest to advertise even in these situations. Paid search ads may give them greater control over how a link is displayed to a user (for example, with text the advertiser has chosen, rather than text that the search engine has retrieved) or guarantee a prominent listing for searches where the advertiser’s URL listing is not always guaranteed to be on top.

Apart from the broader objection to the nature of search advertising, the Roadmap’s authors also object to Google setting an attribution default in its DSP. But advertisers can choose from several different attribution models, not just the default one that the Roadmap objects to, which attributes to search ads the “last click.” Other options include “last non-direct click,” which “ignores direct traffic and attributes 100% of the conversion value to the last channel that the customer clicked through from before buying or converting”; “last Google Ads click”; “First interaction”; and others that give attribution weights according to where and when the customer saw or used the ads during their purchase or conversion “journey.”167F[167] These are precisely the kinds of models that the Roadmap’s authors implicitly believe are more appropriate for campaigns heavy on display advertising. Advertisers can also create their own custom models, and Google has published guides for advertisers to help them choose among the different models.168F[168]

The Roadmap’s objection is thus reduced to being about the choice by Google to use the “last click” attribution model as the default. But some model has to be the default, and “last click” is also the default on, for example, Microsoft Advertising.169F[169] Indeed, according to digital-ad-intermediary company, Outbrain (one of Google’s competitors), it is the most common attribution default across the industry.170F[170] For the Roadmap’s objection to carry any weight, a case based on this claim would need to demonstrate that it was unusual for Google to use “last click” as the default attribution. Even then, given the ease with which advertisers can change the attribution model, the charge would be thin.

3.     ‘Opaque’ pricing

Both the Texas Complaint and the Roadmap allege that Google’s “opaque pricing” constitutes a barrier to entry by impeding “advertisers from switching to a lower-cost for higher-quality” buying tool.171F[171] As the Roadmap puts it, a “new or potential new PAS or DSP cannot credibly claim to be able to undercut the Google products on price if the publishers and advertisers cannot tell how much Google actually is charging.”172F[172] The Texas Complaint further alleges that “Google compounds its exclusionary auction manipulations by purposefully keeping its auction mechanics, terms, and pricing, opaque and ‘nontransparent.’ This makes it nearly impossible for publishers and advertisers to discover Google’s misrepresentations, and even harder for rivals to neutralize or offset.”173F[173] Both the Texas Complaint and the Roadmap also suggest that competition is undermined when publishers and advertisers do not know the fee structure of the intermediaries they are using.

But it is not unusual for businesses’ costs and prices to be private to their competitors, and it is not a barrier to competition. Grocery stores do not need to know how much it cost a farmer to grow an orange or how much their rivals are paying for transportation, unless they are attempting to anticompetitively coordinate their prices; they just need to work to make their own costs as low as possible and to reduce their prices to consumers by as much as possible. Similarly efficient firms are perfectly able to offer competitive prices simply by making the best offer based on their own fundamentals; only less-efficient firms will struggle (as they should).

Along these lines, for competition to work effectively in display advertising, Google’s competitors do not need to know what Google is charging; they need to offer a price and product that is more attractive overall to potential customers than Google’s is. Similarly, a publisher does not need to know how much an advertiser bid to place an ad, nor does the advertiser need to know how much the publisher received to serve the ad. The advertiser’s competition concern is whether an effective ad can be served at a lower price from a different intermediary and the publisher’s competition concern is whether it can earn greater revenues from a different intermediary. One should not be surprised that Google does not reveal information on which competing intermediaries can free ride. Indeed, this is widely considered to be one of the hallmarks of vigorous competition.

Conclusion

As we have argued, many of the most significant claims made against Google’s ad tech products are based on a misunderstanding of U.S. antitrust law, or of the details of the ad tech market itself. Although we cannot be sure how the Texas, et al. v. Google case will develop once the allegations in the Complaint are fleshed out into full arguments, many of its initial claims and assumptions are wrongheaded. Based on the information currently available, if the court rules in favor of these, the result will be to condemn procompetitive conduct and potentially to impose costly, inefficient remedies that function as a drag on innovation.

Legislators, too, who may be concerned about Google’s conduct and tempted to impose regulatory requirements on it and other tech companies should bear in minds the risk of the Nirvana fallacy, in which real-life conduct is compared against a hypothetical “competition maximizing” benchmark, and anything that falls short is deemed problematic and in need of intervention.174F[174] That approach would pervert the incentives of businesses to innovate and compete, and would make an unobtainable “perfect” that exists only in the minds of some economists and lawyers the enemy of a “good” that exists in the market right now.

 

[1] Third Amended Complaint, Texas v. Google, 21-md-3010-PKC (S.D.N.Y. Jan 14, 2022) at 105 (hereinafter, “Texas Complaint”).

[2] See Complaint, United States v. Google LLC, No. 1:20-CV-03010 (D.D.C. Oct. 20, 2020); see also, Complaint, State of Colorado, et al. v. Google LLC, 1:20-CV-03715 (D.D.C. Dec. 17, 2020).

[3] DoJ Expected to File Antitrust Lawsuit Against Google in Weeks—Bloomberg News, U.S. News (Jul. 14, 2022), https://money.usnews.com/investing/news/articles/2022-07-14/doj-expected-to-file-antitrust-lawsuit-against-google-in-weeks-bloomberg-news.

[4] Antitrust: Commission Opens Investigation into Possible Anticompetitive Conduct by Google in the Online Advertising Technology Sector, European Commission (Jun. 22, 2021), https://ec.europa.eu/commission/presscorner/detail/en/ip_21_3143.

[5] Bundeskartellamt Publishes Report on Non-Search Online Advertising for Public Discussion, Bundeskartellamt (Aug. 29, 2022), https://www.bundeskartellamt.de/SharedDocs/Meldung/EN/Pressemitteilungen/2022/29_08_2022_SU_Online_Werbung.html?nn=3599398.

[6] Id.

[7] Online Platforms and Digital Advertising Market Study, U.K. Competition and Markets Authority (Jul. 1, 2020), https://www.gov.uk/cma-cases/online-platforms-and-digital-advertising-market-study (hereinafter “CMA Market Study”); Online Platforms and Digital Advertising, Market Study Final Report, U.K. Competition and Markets Authority (Jul. 1, 2020), https://assets.publishing.service.gov.uk/media/5fa557668fa8f5788db46efc/Final_report_Digital_ALT_TEXT.pdf (hereinafter “CMA Final Report”), at 21 & 37.

[8] See Josh Frydenberg, Competition and Consumer (Price Inquiry—Digital Advertising Services) Direction 2020 (Feb. 10, 2020); Ad Tech Inquiry Issues Paper 5, Australian Competition & Consumer Commission (Mar. 10, 2020); Digital Advertising Services Inquiry, Australian Competition & Consumer Commission (Feb. 26, 2021), https://www.accc.gov.au/focus-areas/inquiries-finalised/digital-advertising-services-inquiry/submissions-to-interim-report.

[9] Investigation of Competition in Digital Markets, Majority Staff Report and Recommendations, Subcommittee on Antitrust, Commercial and Administrative Law of the Committee on the Judiciary (Oct. 4, 2020), available at https://templatelab.com/competition-in-digital-markets/4493-519; Hearing on Stacking the Tech: Has Google Harmed Competition in Online Advertising?, Committee of the Judiciary, Subcommittee on Antitrust, Competition Policy and Consumer Rights (Sep. 15, 2020), https://www.judiciary.senate.gov/meetings/stacking-the-tech-has-google-harmed-competition-in-online-advertising.

[10] Investigation of Competition in Digital Markets, Majority Staff Report and Recommendations, id., at 20.

[11] Competition and Transparency in Digital Advertising Act, S.4258, 117th Congress (2021-2022), https://www.congress.gov/bill/117th-congress/senate-bill/4258/text.

[12] Lee Introduces Digital Advertising Act, Mike Lee US Senator for Utah (May 19, 2022), https://www.lee.senate.gov/2022/5/lee-introduces-digital-advertising-act.

[13] Support Mounts for Lee’s Digital Advertising Act, Mike Lee US Senator for Utah (May 27, 2022), https://www.lee.senate.gov/2022/5/support-mounts-for-lee-s-digital-advertising-act.

[14] Online Platforms and Digital Advertising Market Study Interim Report, U.K. Competition and Markets Authority (Dec. 18, 2019), https://assets.publishing.service.gov.uk/media/5ed0f75bd3bf7f4602e98330/Interim_report_—_web.pdf (hereinafter, “CMA Interim Report”).

[15] Fiona M. Scott Morton & David C. Dinielli, Roadmap for a Digital Advertising Monopolization Case Against Google, Omidyar Network (May 2020), https://omidyar.com/wp-content/uploads/2020/09/Roadmap-for-a-Case-Against-Google.pdf (hereinafter, “Roadmap” or “Omidyar Roadmap”). One of the Roadmap’s authors testified at a Senate hearing on the display-advertising market, and the report has been widely cited. See, e.g., Gilad Edelman, Here’s What an Antitrust Case Against Google Might Look Like, Wired (May 18, 2020), https://www.wired.com/story/antitrust-case-against-google-roadmap-paper.

[16] Damien Geradin & Dimitrios Katsifis, An EU Competition Law Analysis of Online Display Advertising in the Programmatic Age, 15 Eur. Comp. J. 55 (2019); Damien Geradin & Dimitrios Katsifis, “Trust Me, I’m Fair”: Analysing Google’s Latest Practices in Ad Tech from the Perspective of EU Competition Law, 16 Eur. Comp. J. 11 (2020); Damien Geradin & Dimitrios Katsifis, Online Platforms and Digital Advertising Market Study: Observations on CMA’s Interim Report, TILEC Discussion Paper No. DP2020-044 (Feb. 13, 2020), https://ssrn.com/abstract=3537864; Damien Geradin & Dimitrios Katsifis, Competition in Ad Tech: A Response to Google, TILEC Discussion Paper No. DP2020-038 (Jun. 3, 2020), https://ssrn.com/abstract=3617839.

[17] The markets alleged in the Texas Complaint involve (1) publisher ad servers, (2) ad exchanges, (3) ad-buying tools for large advertisers, (4) ad-buying tools for small advertisers, (5) in-app mediation tools, and (6) in-app networks. The complaint does not relate to other forms of advertising on the Internet, such as targeted text-based ads sold by search engines, video ads that run before or during video content, or shareable ads on social media platforms.

[18] This section is distilled from our much longer discussion of the broader market surrounding digital advertising. See Eric Fruits, Geoffrey A. Manne & Lazar Radic, Relevant Market in the Google AdTech Case, ICLE Issue Brief 2022-06-01 (2022), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4125569.

[19] Benedict Evans, News by the Ton: 75 Years of US Advertising (Jun. 15, 2020), https://www.ben-evans.com/benedictevans/2020/6/14/75-years-of-us-advertising; Benedict Evans, TV, Merchant Media and the Unbundling of Advertising (Mar. 18, 2022), https://www.ben-evans.com/benedictevans/2022/3/18/unbundling-advertising.

[20] See Fruits, Manne & Radic, supra note 18.

[21] Michael Schneider & Kate Aurthur, R.I.P. Cable TV: Why Hollywood Is Slowly Killing Its Biggest Moneymaker, Variety (Jul. 21, 2020), https://variety.com/2020/tv/news/cable-tv-decline-streaming-cord-cutting-1234710007 (“[B]asic cable feasted on a dual revenue stream of subscriber fees and advertising revenue. But that gravy train started going off the rails when the streaming services arrived.”).

[22] At the same time, as Benedict Evans notes, not all digital advertising is drawn from offline sources: “[I]f you talk to people at both Google and Facebook and in the agency world, you’ll hear that a lot of the money spent on Google and Facebook is money that was never spent on traditional advertising—it’s coming from SMEs [small and medium enterprises] and local businesses that might have spent in classified at most but probably wouldn’t have done even that.” Evans, News by the Ton, supra note 19 (emphasis in original).

[23] See Xi He, Rigoberto Lopez & Yizao Liu, Are Online and Offline Advertising Substitutes or Complements? Evidence from U.S. Food Industries, 15 J. Agricultural & Food Indus. Org. 1 (2017).

[24] David Bardey, Jorge Tovar & Nicolas Santos, Characterization of the Relevant Market in the Media Industry: Some New Evidence, Toulouse School of Economics Working Paper 16-719 (2016), https://www.tse-fr.eu/publications/characterization-relevant-market-media-industry-some-new-evidence (“The results show substitution and complementary patterns across certain media outlets. An increase in price for advertising in radio, for instance, leads to higher demand for newspapers and outdoors. Similarly, complementarity relationships between media outlets are observed, suggesting that advertising across the various media platforms is, overall, interwoven.”).

[25] David S. Evans, The Online Advertising Industry: Economics, Evolution, and Privacy, 23 J. Econ. Persp. 37, 49 (2009).

[26] Avi Goldfarb & Catherine Tucker, Search Engine Advertising: Channel Substitution When Pricing Ads to Context, 57 Management Sci. 458 (2011) (The authors find that the price of “ambulance chaser” lawyer ads was significantly more expensive in states prohibiting direct-mail solicitation by attorneys. This leads them to conclude that “online advertising substitutes for online advertising”).

[27] Avi Goldfarb & Catherine Tucker, Substitution Between Offline and Online Advertising Markets, 7 J. Competition L. & Econ. 37, 43 (2011).

[28] Daniel S. Levine, Ad-Supported Cyber-Magazines to Launch on Internet, Adweek (Sep. 10, 1993).

[29] Id.

[30] Brian Morrissey, How the Banner Ad Was Born, Digiday (Apr. 12, 2013), https://digiday.com/marketing/how-the-banner-ad-was-born.

[31] Chris Lapham, AOL and GNN Partner to Build Launch Pad, CMC Magazine (Jul. 1, 1995), https://www.december.com/cmc/mag/1995/jul/cutting.html.

[32] Kim Cleland, Poppe Creates Web Net, Advertising Age (Oct. 30, 1995).

[33] Id.

[34] The History of Online Advertising, OKO Ad Management (Jul. 19, 2019), https://oko.uk/blog/the-history-of-online-advertising.

[35] The company changed its name to Overture, which was acquired by Yahoo! in 2003.

[36] GoTo.com Announces First Round of Financing, Totaling More Than $ 6 Million, Led by Draper Fisher Jurvetson, Business Wire (May 19, 1998), https://www.internetnews.com/marketing/goto-com-raises-6-million-in-first-round-of-financing.

[37] Will Oremus, Google’s Big Break, Slate (Oct. 13, 2013), https://slate.com/business/2013/10/googles-big-break-how-bill-gross-goto-com-inspired-the-adwords-business-model.html.

[38] Id.

[39] Dean Schmid, The History of Display Advertising: Everything You Need to Know, DisruptorDaily.com (Aug. 14, 2017), https://www.disruptordaily.com/the-history-of-display-advertising-everything-you-need-to-know.

[40] Julia Scheeres, Death of Banner Ads Exaggerated, Wired (Jan. 26, 2001), https://www.wired.com/2001/01/death-of-banner-ads-exaggerated.

[41] Id.

[42] Breaking News, AdAge (Oct. 23, 2000).

[43] Mark Evans, Investors Leap off Overture Roller Coaster: Rival Google Elbows In, National Post (Feb. 21, 2002).

[44] Oremus, supra note 37.

[45] Google Grabs Applied Semantics, EuropeMedia (Apr. 25, 2003); Google Expands Advertising Monetization Program for Websites, Google Press Release (Jun. 18, 2003), http://googlepress.blogspot.com/2003/06/google-expands-advertising-monetization.html.

[46] Dean Schmid, The History of Display Advertising: Everything You Need to Know, DisruptorDaily (Aug. 14, 2017), https://www.disruptordaily.com/the-history-of-display-advertising-everything-you-need-to-know.

[47] Kate Walsh, Search Marketing: Understanding the Basics, B2B Marketing Magazine (March 2006).

[48] Louise Story & Miguel Helft, Google Buys DoubleClick for $3.1 Billion, The New York Times (Apr. 14, 2007), https://www.nytimes.com/2007/04/14/technology/14DoubleClick.html.

[49] Sarah Sluis, The Year Header Bidding Went Mainstream, AdExchanger (Dec. 27, 2016); Townhall Media Selects OpenX for Patent-Pending Header Bidding Solution, BusinessWire (Sep. 18, 2015), https://www.businesswire.com/news/home/20150918005110/en/Townhall-Media-Selects-OpenX-for-Patent-Pending-Header-Bidding-Solution.

[50] As the name suggests, ad waterfalls enable publishers to sell their inventory seriatim, beginning with premium, direct sales and flowing through the most historically profitable ad servers in succession to unload unsold inventory before offering its remnant inventory in the open display channel. See, e.g., Maciej Zawadzinski, What Is Waterfalling and How Does it Work?, Clearcode (Aug. 20, 2021), https://clearcode.cc/blog/what-is-waterfalling.

[51] See, e.g., Header Bidding, OKO Ad Management, https://oko.uk/topic/header-bidding (retrieved July 27, 2022).

[52] Client-side header bidding is so-named because it operates via a small piece of java script embedded in the header of a publisher’s website and executed within the user’s browser (i.e., client). See, e.g., Maciej Zawadzinski, What Is Header Bidding and How Does it Work?, Clearcode (Aug. 20, 2021), https://clearcode.cc/blog/what-is-header-bidding.

[53] Header Bidding Facts and Statistics 2021, Automatad (Jun. 27, 2021), https://headerbidding.co/header-bidding-statistics. Today, 70% of the top 10,000 U.S. publishers use header bidding. See Header Bidding (HBIX) Tracker, kevel (retrieved Nov. 1, 2022), https://www.kevel.com/hbix.

[54] See, e.g., CMA Final Report, supra note 7, at Appendix M, ¶ 33.

[55] See, e.g., Vishveshwar Jatain, Header Bidding Integrations: Client Vs. Server-Side, Explained, Blockthrough (Apr. 15, 2021), https://blockthrough.com/blog/header-bidding-integrations-client-vs-server-side-explained.

[56] IAB and PwC, IAB Internet Advertising Revenue Report, 2010 Full Year Results (Apr. 2011), available at https://www.iab.com/wp-content/uploads/2015/05/IAB_Full_year_2010_0413_Final.pdf; Megan Graham, Digital Ad Revenue Jumped 35% in the U.S. Last Year, Biggest Gain Since 2006, Wall Street Journal (Apr. 12, 2022), https://www.wsj.com/articles/digital-ad-revenue-jumped-35-in-the-u-s-last-year-biggest-gain-since-2006-11649759401.

[57] Producer Price Index by Commodity: Advertising Space and Time Sales: Internet Advertising Sales, Excluding Internet Advertising Sold by Print Publishers, U.S. Bureau of Labor Statistics, https://fred.stlouisfed.org/series/WPU365; Producer Price Index, December 2009—February 2021, U.S. Bureau of Labor Statistics, https://fred.stlouisfed.org/graph/?g=vtTd.

[58] Scott Morton & Dinielli, supra note 15, at 9.

[59] Importantly, however, network effects are not monolithic; nor do they increase forever. For different types of networks at different points in their growth, adding more users might not increase the value of the platform and could even reduce the platform’s benefits. See, e.g., D’Arcy Coolican & Li Jin, The Dynamics of Network Effects, Andreesen Horowitz (Dec. 13, 2018), https://a16z.com/2018/12/13/network-effects-dynamics-in-practice.

[60] See Stan J. Liebowitz & Stephen E. Margolis, Network Externality: An Uncommon Tragedy, 8 J. Econ. Persp. 133 (1994).

[61] Catherine Tucker, Digital Data, Platforms and the Usual [Antitrust] Suspects: Network Effects, Switching Costs, Essential Facility, 54 Rev. Indus. Org. 683, 686 (2019).

[62] See, e.g., David S. Evans, Economics of Vertical Restraints for Multi-Sided Platforms, University of Chicago Institute for Law & Economics Olin Research Paper No. 626 (Jan. 2, 2013), https://ssrn.com/abstract=2195778.

[63] See, e.g., Stylianos Despotakis, R. Ravi & Amin Sayedi, First-Price Auctions in Online Display Advertising, 58 J. Marketing Research 888 (2021). See also Display Advertising Switched to First-Price Auctions After Adoption of Header Bidding, New Study Finds, Tepper School of Business (Apr. 22, 2020), https://www.cmu.edu/tepper/news/stories/2020/april/display-advertisting-research-ravi.html.

[64]  Jonathan Levin, Auction Theory (Oct. 2004), available at https://web.stanford.edu/~jdlevin/Econ%20286/Auctions.pdf.

[65] Maciej Zawadzi?ski, How Do First-Price and Second-Price Auctions Work in Online Advertising?, Clearcode (Aug. 12, 2021), https://clearcode.cc/blog/first-price-second-price-auction.

[66] Id.

[67] Texas Complaint, supra note 1, at ¶ 351 (“Overall, the lack of transparency prevents more efficient competition that would drive greater innovation, increase the quality of intermediary services, increase output, and create downward pricing pressure on intermediary fees.”); Scott Morton & Dinielli, supra note 15, at 18 (“Based on the public facts known at the moment, however, it does not seem plausible that the incremental efficiencies created by the conduct described here could outweigh all the harms to competition resulting from this broad pattern of behaviors.”); Scott Morton & Dinielli, supra note 15, at 38 (“It also is true that Google has allowed some rivals to survive (although not necessarily to thrive). It is possible that Google adopted a strategy of incomplete foreclosure specifically so that it can paint an illusion of healthy competition when the reality is quite different. Indeed, to the extent Google has adopted ‘pro-competitive’ concessions, the narrative here demonstrates that they simply have not succeeded in addressing the harms or lowering the barriers to entry.”).

[68] Id. at 3 (“It is clear even to us as lay people that there are less anticompetitive ways of delivering effective digital advertising—and thereby preserving the substantial benefits from this technology—than those employed by Google.”).

[69] United States v. Aluminum Co. of America, 148 F.2d 416, 430 (2^nd Cir. 1945) (Learned Hand, J.) (emphasis added).

[70] Cal. Computer Prods., Inc. v. Int’l Bus. Machine Corp., 613 F.2d 727, 744 (9th Cir. 1979) (“IBM, assuming it was a monopolist, had the right to redesign its products to make them more attractive to buyers whether by reason of lower manufacturing cost and price or improved performance. It was under no duty to help [its competitors] survive or expand.”).

[71] Scott Morton & Dinielli, supra note 15 at 18.

[72] Texas Second Amended Complaint at ¶ 113.

[73] See Verizon Commc’ns Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S. 398, 408 (2004). The exception—”at or near the outer boundary of § 2 liability” (id. at 409)—is the extremely narrow case in which a monopolist (i) sacrifices profits, by (ii) terminating a prior course of dealing, (iii) for no purpose except to harm competition. See Novell v. Microsoft, 731 F.3d 1064, 1074-75 (10th Cir. 2013) (Gorsuch, J.) (holding that a refusal-to-deal claim requires terminating “a preexisting voluntary” course of dealing where the “monopolist decided to forsake short-term profits,” and “the monopolist’s conduct” is “irrational but for its anticompetitive effect”).

[74] Opinion and Order, Texas, et al. v. Google, 21-md-3010-PKC (S.D.N.Y, Sep. 13, 2022) (citing Charych v. Siriusware, Inc., 790 Fed. App’x 299, 302 (2^nd Cir. 2019)).

[75] United States v. Colgate & Co., 250 U.S. 300, 307 (1919).

[76] Trinko, 540 U.S. at 407-08

[77] See Brunswick Corp. v. Pueblo Bowl-O-Mat, Inc., 429 U.S. 477, 488 (1977) (“The antitrust laws, however, were enacted for ‘the protection of competition not competitors.’”) (quoting Brown Shoe Co. v. United States, 370 U.S. 294, 320 (1962)).

[78] See Ohio v. American Express, 138 U.S. 2274, 2285 (2018) (“Due to indirect network effects, two-sided platforms cannot raise prices on one side without risking a feedback loop of declining demand…. Price increases on one side of the platform [] do not suggest anticompetitive effects without some evidence that they have increased the overall cost of the platform’s services.”).

[79] Geoffrey A. Manne, In Defence of the Supreme Court’s ‘Single Market’ Definition in Ohio v American Express, 7 J. Antitrust Enforcement 104, 111 (2019) (quoting Brief for Amici Curiae Antitrust Law & Economics Scholars in Support of Respondents at 19, Ohio v. American Express, 138 U.S. 2274 (2018) (No. 16-1454) and United States, et al. v. American Express, 838 F.3d 179, 198 (2^nd Cir. 2016)).

[80] Among innumerable examples, see Texas Complaint, supra note 1, at ¶ 297 (“Google’s harm to the competitive process has harmed customers in this market, i.e., online publishers.”). Notably, the Texas Complaint does, in places, recognize that identifying the incidence of benefits and harms in multisided markets is complex—it just fails to carry its analysis to its logical conclusion. Thus, in ¶157 the Complaint notes that “[t]he higher advertising revenue publishers make from exchanges permits publishers to offer consumers better quality content and lower-priced or free access to their content.” (Emphasis added). Undoubtedly, this is true. But if it is correct, then it must also be correct that, at the same time, the correspondingly higher prices advertisers pay for advertising through exchanges limits their ability to provide marketing benefits directly to consumers and may increase the price to consumers of the advertised goods. It is an empirical question which effect is larger, but the mere possibility that one set of consumers could benefit from a different arrangement is insufficient on its own to identify harm when another set of consumers would be harmed by it.

[81] See Harold Demsetz, Information and Efficiency: Another Viewpoint, 12 J. L. & Econ. 1, 1-2 (1969) (“In practice, those who adopt the nirvana viewpoint seek to discover discrepancies between the ideal and the real and if discrepancies are found, they deduce that the real is inefficient…. The nirvana approach is… susceptible… to committing three logical fallacies—the grass is always greener fallacy, the fallacy of the free lunch, and the people could be different fallacy.”) (emphasis in original).

[82] See, generally, Thomas Nachbar, Less Restrictive Alternatives and the Ancillary Restraints Doctrine, Virginia Law and Economics Research Paper No. 2020-18 (2021) (forthcoming U. Seattle L. Rev.) at 57-8, available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3723807 (“The more general risk to tech markets comes from the intangible nature of the products and services they produce. Although many of the cases cited for less restrictive alternatives are horizontal cases, it is the vertical context (which normally receives more permissive antitrust review) in which less restrictive alternatives present the greatest likelihood of destabilizing current law because of the difficulty of specifying what is and is not less restrictive with regard to the intangible products produced by today’s ‘big tech’ economy. To the extent that less restrictive alternatives present problems of incrementalism, those problems will be exacerbated in the ‘big tech’ markets.”).

[83] See Geoffrey A. Manne, Error Costs in Digital Markets, Global Antitrust Institute Report on the Digital Economy (Joshua D. Wright & Douglas H. Ginsburg, eds., 2020) 33, 76, available at https://gaidigitalreport.com/wp-content/uploads/2020/11/Manne-Error-Costs-in-Digital-Markets.pdf (“The concern with error costs is especially high in dynamic markets in which it is difficult to discern the real competitive effects of a firm’s conduct from observation alone. And for several reasons, antitrust decision-making in the context of innovation tends much more readily toward distrust of novel behavior, thus exacerbating the risk and cost of over-enforcement.”).

[84] Among many other examples, see Texas Second Amended Complaint at ¶138 (“Then, through Dynamic Allocation, Google’s ad server passed inside information to Google’s exchange and permitted Google’s exchange to purchase valuable impressions at artificially depressed prices. Competing exchanges were deprived of the opportunity to compete for inventory and left with the low-value impressions passed over by Google’s exchange.”); Omidyar Roadmap, supra note 15, at 20 (“[A]fter purchasing DoubleClick, which became its publisher ad server, Google apparently lowered its prices to publishers by a factor of ten, at least according to one publisher’s account related to the CMA. Low prices for this service can force rivals to depart, thereby directly reducing competition.”).

[85] Barry Wright Corp. v. ITT Grinnell Corp., 724 F.2d. 227, 235 (1^st Cir. 1983) (Breyer, C.J.).

[86] Allied Orthopedic Appliances, Inc. v. Tyco Health Care Grp. LP, 592 F.3d 991, 1000 (9^th Cir. 2010).

[87] Berkey Photo, Inc. v. Eastman Kodak Co., 603 F.2d 263, 287 (2^nd Cir. 1979). See also, Manne & Wright, Innovation and the Limits of Antitrust, 6 J. Comp. L. & Econ. 153–202 (March 2010), https://doi.org/10.1093/joclec/nhp032.

[88] Allied Orthopedic, 592 F.3d at 999-1000; see also, California Computers Prods. v. IBM, 613 F.2d 727, 744 (9^th Cir. 1979); Foremost Pro Color, Inc. v. Eastman Kodak Co., 703 F.2d 534, 543-45 (9^th Cir. 1983).

[89] Foremost Pro Color, 703 F.2d at 543.

[90] Allied Orthopedic, 592 F.3d at 1000.

[91] Verizon Commc’ns, Inc. v. Law Offices of Curtis V. Trinko, 540 U.S. 398, 400-41 (2004).

[92] Trinko, 540 U.S. at 415-16.

[93] Id.; see also New York Merc. Exch., Inc. v. Intercontinental Exch. Inc., 323 F.Supp.2d 559 (S.D.N.Y. 2004) (dismissing Section 2 claim and reiterating limited exceptions in which forced sharing is appropriate).

[94] See, e.g., Jack Walters & Sons Corp. v. Morton Bldg., Inc., 737 F.2d 698, 710 (7^th Cir. 1984) (“We just said that vertical integration is not an improper objective. But this puts the matter too tepidly; vertical integration usually is procompetitive. If there are cost savings from bringing into the firm a function formerly performed outside it, the firm will be made a more effective competitor.”). There is a robust body of empirical research indicating that vertical integration is generally procompetitive or benign. For a summary of the leading meta-studies by DOJ and FTC economists and others, see Koren W. Wong-Ervin, Antitrust Analysis of Vertical Mergers: Recent Developments and Economic Teachings, The Antitrust Source (February 2019), https://www.americanbar.org/content/dam/aba/publishing/antitrust_source/2018-2019/atsource-february2019/feb19_wong_ervin_2_18f.pdf. See also, Francine Lafontaine & Margaret Slade, Vertical Integration and Firm Boundaries: The Evidence, 45 J. Econ. Lit. 677 (2007) (“In spite of the lack of unified theory, overall a fairly clear empirical picture emerges. The data appear to be telling us that efficiency considerations overwhelm anticompetitive motives in most contexts. Furthermore, even when we limit attention to natural monopolies or tight oligopolies, the evidence of anticompetitive harm is not strong.”). See also, generally, Geoffrey A. Manne, Kristian Stout & Eric Fruits, The Fatal Economic Flaws of the Contemporary Campaign Against Vertical Integration, 68 Kansas L. Rev. 923 (2020).

[95] Scott Morton & Dinielli, supra note 15, at 17.

[96] See, e.g., Port Dock & Stone Corp. v. Oldcastle Ne., Inc., 507 F.3d 117, 123-25 (2^nd Cir. 2007) (affirming dismissal of a Section 2 claim and finding that even a monopolist’s “vertical expansion into another level of the same product market will ordinarily be for the purpose of increasing its efficiency, which is a prototypical valid business purpose”). Moreover, single-firm conduct that supposedly projects power into another market, even through anticompetitive means, does not violate Sherman Act Section 2 unless the practices threaten monopoly power in that distinct second market. Harming competition is not enough. See Trinko, 540 U.S. at 415 n.4 (citing Spectrum Sports, Inc. v. McQuillan, 506 U.S. 447, 459 (1993)).

[97] Thus, the Omidyar Roadmap condemns Google’s supposed integration of data “to maximize the effectiveness and precision of ad targeting and attribution and thereby the value of an ad,” Scott Morton & Dinielli, supra note 15, at 20, even though the conduct makes Google’s offering to advertisers more attractive.

[98] Id. at 18-19.

[99] CMA Interim Report, supra note 14, at ¶ 5.89.

[100] Texas Second Amended Complaint at ¶ 113.

[101] Scott Morton & Dinielli, supra note 15, at 22.

[102] Texas Second Amended Complaint at ¶¶ 284-91 (“Cutting off access to YouTube foreclosed competition in the ad buying tool markets and protected Google’s market power in these markets. Many DSPs stopped growing, many others went out of business, and the market overall has been closed to entry.”).

[103] Scott Morton & Dinielli, supra note 15, at 22.

[104] See, e.g., Ryan Joe, The Big Story: Call of the Peacock, AdExchanger (Jan. 22, 2020) at 31:05-31:26, https://www.adexchanger.com/podcast/the-big-story/the-big-story-call-of-the-peacock (indicating that NBC’s Peacock streaming service will have only direct sales when it launches); Kevin Weiss, What Is the Amazon Demand Side Platform (DSP)?, Amplio (July 2019), https://www.ampliodigital.com/blog/what-is-the-amazon-demand-side-platform-dsp (“Amazon DSP is the only way to access advertising inventory exclusively available on Amazon’s collection of owned online properties and devices like: Kindle; Fire TV; IMDb; Amazon Owned & Operated properties”); Tim Cross, Xandr Launches New Demand-Side Platform ‘Xandr Invest’, VideoAdNews (Jun. 10, 2019) https://videoadnews.com/2019/06/10/xandr-launches-new-demand-side-platform-xandr-invest (“Xandr [AT&T-Time Warner’s ad tech division] has announced it will be the exclusive source of inventory from Community, its recently announced video marketplace which includes content from various WarnerMedia brands as well as Vice, Hearst Magazines, Newsy, Philo, Tubi and XUMO.”).

[105] Neal Mohan, Focusing Investments to Improve Buying on YouTube, Google (Aug. 6, 2015) https://doubleclick-advertisers.googleblog.com/2015 (“To continue improving the YouTube advertising experience for as many of our clients as possible, we’ll be focusing our future development efforts on the formats and channels used by most of our partners. To enable that, as of the end of the year, we’ll no longer support the small amount of YouTube buying happening on the DoubleClick Ad Exchange.”); see also, Kelly Liyakasa, Google to Yank YouTube Inventory out of AdX by Year’s End, AdExchanger (Aug. 6, 2015), https://www.adexchanger.com/ad-exchange-news/google-to-yank-youtube-inventory-out-of-adx-by-years-end.

[106] Liyakasa, id.

[107] See Lara O’Reilly, Ad Tech Company The Trade Desk Goes Public at $28.75 Per Share—A Huge Pop on its $18 Price Target, Business Insider (Sep. 21, 2016), https://www.businessinsider.com/the-trade-desk-ipo-2016-9; Trey Titone, The Bill That Could Break Up Google and Shake Up Ad Tech, Ad Tech Explained (May 23, 2022), https://adtechexplained.com/competition-and-transparency-in-digital-advertising-act-ctda; Trade Desk Market Cap, YCharts, https://ycharts.com/companies/TTD/market_cap.

[108] Scott Morton & Dinielli, supra note 15, at 16 n.70 (identifying AppNexus as a “vigorous competitor to Google”).

[109] Id. at 2, 28-29.

[110] Annual Report (Form 10-K) for Year Ending December 31, 2021, Alphabet Inc. (Feb. 02, 2022), https://www.sec.gov/ix?doc=/Archives/edgar/data/0001652044/000165204422000019/goog-20211231.htm.

[111] Id.

[112] Id.

[113] Id.

[114] Id.

[115] Rachel Kaser, YouTube Claims to Share Billions in Ad Money with Creators, Unlike Instagram, The Next Web (Feb. 5, 2020), https://thenextweb.com/facebook/2020/02/05/youtube-claims-share-billions-ad-money-creators-unlike-instagram.

[116] Texas Complaint, supra note 1, at ¶¶ 61, 156, 253, 288.

[117] Id. ¶ 157.

[118] Id. ¶ 21.

[119] Scott Morton & Dinielli, supra note 15, at 14 (“The CMA estimates Google’s take rate, or price, at 40%, which it deems a supra-competitive price for the services provided by the Google-controlled players in ad tech stack. A recent study by the Incorporated Society of British Advertisers (ISBA) found that publishers received 51% of the price, while the amount they could track going to intermediaries was 34%. The study could not find where the remaining 15% of the price went. As we will describe below, Google has such a dominant position across all elements of the ecosystem, it seems likely that these missing funds are accruing to Google at least in part, which would support the CMA’s findings.”).

[120] Geradin & Katsifis, “Trust Me, I’m Fair,” supra note 16 (“[L]ack of competition across the ad tech chain enables Google to exploit advertisers and publishers by charging hidden fees for ad intermediation on top of its disclosed commission…. Unfortunately, we conclude that Google’s latest switch does nothing to increase auction transparency. Worse, it seems to strengthen Google’s ability to extract hidden margins from its customers, while undermining the competitive pressure exercised by header bidding.”).

[121] CMA Final Report, supra note 7, at 275 (emphasis added).

[122] Scott Morton & Dinielli, supra note 15, at 20.

[123] See, e.g., Jean-Charles Rochet & Jean Tirole, Platform Competition in Two-Sided Markets, 1 J. Eur. Econ. Ass’n 990 (2003); Bruno Jullien, Price Skewness and Competition in Multi-Sided Markets, IDEI Working Paper 504 (March 2008), available at https://core.ac.uk/download/pdf/6375977.pdf.

[124] See Joshua. D. Wright & John. M. Yun, Burdens and Balancing in Multisided Markets: The First Principles Approach of Ohio v. American Express, 54 Rev. Industrial Organization 717 (2019); Manne, In Defence of the Supreme Court’s ‘Single Market’ Definition in Ohio v American Express, supra note 79.

[125] As described here, true pricing is theoretically possible but difficult in practice: “To successfully engage in predatory pricing means taking enormous and rising losses that grow for the ‘predatory’ firm as customers switch to it from its competitor. And once the rival firm has exited the market, if the predatory firm raises prices above average cost (i.e., to recoup its losses), there is no guarantee that a new competitor will not enter the market selling at the previously competitive price. And the competing firm can either shut down temporarily or, in some cases, just buy up the ‘predatory’ firm’s discounted goods to resell later.” Sam Bowman, Buck’s “Third Way”: A Different Road to the Same Destination, Truth on the Market (Oct. 27, 2020), https://truthonthemarket.com/2020/10/27/bucks-third-way-a-different-road-to-the-same-destination.

[126] Scott Morton & Dinielli, supra note 15, at 20.

[127] See, e.g., Barry Wright, 724 F.2d at 234-35.

[128] CMA Interim Report, supra note 14, at Appendix H, ¶ 194.

[129] Texas Complaint, supra note 1, at ¶ 477.

[130] Id. at ¶¶ 473-476.

[131] Scott Morton & Dinielli, supra note 15, at 28.

[132] Id. at 28.

[133] As one study on the effects of GDPR (in this case, on app development) notes, “While 42.1 percent of EU-developed apps exit in the year following GDPR, the analogous ?gure averages between 37.7 and 50 percent in the other six countries, con?rming the di?culty in ?nding an untreated part of the world.” Rebecca Janßen, Reinhold Kesler, Michael E. Kummer, & Joel Waldfogel, GDPR and the Lost Generation of Innovative Apps, NBER Working Paper 30028 (May 2022) at 19-20, available at https://www.nber.org/papers/w30028.

[134] Julie Brill, Microsoft Will Honor California’s New Privacy Rights Throughout the United States, Microsoft Blog (Nov. 11, 2019), https://blogs.microsoft.com/on-the-issues/2019/11/11/microsoft-california-privacy-rights.

[135] Nick Statt, Apple Updates Safari’s Anti-Tracking with Full Third-Party Cookie Blocking, The Verge (Mar. 24, 2020), https://www.theverge.com/2020/3/24/21192830/apple-safari-intelligent-tracking-privacy-full-third-party-cookie-blocking.

[136] Dieter Bohn, Google to “Phase Out” Third-Party Cookies in Chrome, but not for Two Years, The Verge (Jan. 24, 2020), https://www.theverge.com/2020/1/14/21064698/google-third-party-cookies-chrome-two-years-privacy-safari-firefox.

[137] Scott Morton & Dinielli, supra note 15, at 30.

[138] Id. at 20.

[139] Texas Complaint, supra note 1, at ¶¶ 407-408. See also, Scott Morton & Dinielli, supra note 15, at 26.

[140] See David Besbris, Introducing the Accelerated Mobile Pages Project, for a Faster, Open Mobile Web, Google (Oct. 7, 2015), https://blog.google/products/search/introducing-accelerated-mobile-pages/.

[141] See Automated Team, Header Bidding on AMP—A Complete Guide, Automated (Jan. 10, 2020), https://headerbidding.co/header-bidding-amp.

[142] Scott Morton & Dinielli, supra note 15, at 26.

[143] Id. at 17.

[144] See William J. Baumol, Contestable Markets: An Uprising in the Theory of Industry Structure, 72 Am. Econ. Rev. 1, 14 (1982) (“In the limit, when entry and exit are completely free, efficient incumbent monopolists and oligopolists may in fact be able to prevent entry. But they can do so only by behaving virtuously, that is, by offering to consumers the benefits which competition would otherwise bring. For every deviation from good behavior instantly makes them vulnerable to hit-and-run entry.”).

[145] Scott Morton & Dinielli, supra note 15, at 15.

[146] See, generally, William J. Baumol, John C. Panzar, & Robert D. Willig, Contestable Markets and the Theory of Industry Structure (1982).

[147] United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001) (“Because a firm cannot possess monopoly power in a market unless that market is also protected by significant barriers to entry… it follows that a firm cannot threaten to achieve monopoly power in a market unless that market is, or will be, similarly protected.”).

[148] Scott Morton & Dinielli, supra note 15, at 15.

[149] Id.

[150] See CMA Final Report, supra note 7, at 252–55 for a discussion of barriers to entry.

[151] See Texas Complaint, supra note 1, at ¶ 127: In addition to these barriers, Google’s own anticompetitive conduct imposes additional barriers to entry and expansion. As addressed below in Section VII.A, from 2010 to present, Google has tied its ad server to its ad exchange, requiring publishers to use Google’s ad server in order to receive live, competitive bids from Google’s ad exchange. This tie effectively forces almost every large publisher to use Google’s ad server. And because it is difficult-to-impossible for a publisher to use multiple ad servers simultaneously, requiring publishers to use Google’s ad server effectively prohibits them from using a competitor’s ad server. Google’s anticompetitive conduct creates an unnatural and nearly insurmountable barrier to entry.

[152] An earlier version of the Texas Complaint did make assertions regarding Google’s abuse of monopoly power through the “use [of] its data advantages to trade on inside information” (Texas Second Amended Complaint at ¶ 311), by which the state plaintiffs may mean (or have meant) to encompass location data, among other things.

[153] Scott Morton & Dinielli, supra note 15, at 15.

[154] CMA Interim Report, supra note 14, at 189 (emphasis added).

[155] Id., at ¶ 3.71 (“Microsoft suggested that accessing at-scale location data from user devices is a critical input to providing relevant, localized results. It indicated its belief that Google has unique advantages in this area, due to the location data that it receives from the Android operating system and the location data it receives when users access Google Search or other apps like Google Maps/Waze.”).

[156] CMA Final Report, supra note 7, at ¶ 5.268.

[157] Scott Morton & Dinielli, supra note 15, at 15.

[158] To be sure, location data can be helpful in assessing the efficacy of advertising by, for example, enabling an advertiser to better evaluate whether an advertisement led users to go to the advertiser’s physical location. But this function hardly seems necessary to a well-functioning market, and other sources of such information (e.g., questionnaires) are available.

[159] Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the Protection of Natural Persons with Regard to the Processing of Personal Data and on the Free Movement of Such Data, and Repealing Directive 95/46/EC (General Data Protection Regulation), OJ L 119, 4.5, 2016; See, e.g., Bert Peeters, Processing of Location Data: Navigating the EU Data Protection Framework, CiTiP Blog (Feb. 4, 2021), https://www.law.kuleuven.be/citip/blog/processing-of-location-data-navigating-the-eu-data-protection-framework. (“The general understanding seems to be that, while European law does not qualify location data as a ‘special category’ of data under article 9 of the GDPR, location data should for all intents and purposes be treated with the utmost of care.”).

[160] Avi Goldfarb & Catherine Tucker, Privacy Regulation and Online Advertising, 57 Mgmt. Sci. 57, 57 (2011).

[161] Nils Wernerfelt, Anna Tuchman, Bradley Shapiro, & Robert Moakler, Estimating the Value of Offsite Data to Advertisers on Meta, University of Chicago, Becker Friedman Institute for Economics Working Paper No. 114 (August 22, 2022) at 1, available at https://ssrn.com/abstract=4198438 (“Taken together, our results suggest a substantial benefit of offsite data across a wide range of advertisers, an important input into policy in this space.”).

[162] Michal S. Gal & Oshrit Aviv, The Competitive Effects of the GDPR, 16 J. Competition & Econ. 349 (May 18, 2020). See also James Campbell, Avi Goldfarb, & Catherine Tucker, Privacy Regulation and Market Structure, 24 J. Econ. & Mgmt. Strategy 47, 68 (2015) (“[A] potential risk in privacy regulation is the entrenchment of the existing incumbent firms and a consequent reduction in the incentives to invest in quality. These incentives are stronger when firms have little consumer-facing price flexibility, as is the case in online media.”).

[163] Gal & Aviv, id. at 16.

[164] See, e.g., Cheok Lup, Explaining Marketing Attribution Models [Scenario Example], tinkerEdge (Nov. 12, 2015), https://www.tinkeredge.com/blog/web-analytics/explaining-marketing-attribution-models. (“On Day #1: User wants to purchase a coffee table for his new house, and perform a keyword search on Google. He clicks on one of the organic listings on Google Search Engine Result Page (SERP) to land onto Overstock.com. On Day # 2: He continues his search for his coffee table, and clicks on one of the PPC ads on Google SERP to land onto Overstock.com again. He subscribes to the email newsletter this time. On Day #3: He receives an eDM [electronic direct mail] from Overstock.com with a promotional offer of 30% discount sale, and clicks the “Buy Now” button from the eDM to enter the website. Unable to resist the discount offer, he decides to make a purchase of the furniture from the website.”). Attribution metrics determine which channel gets credit for the ultimate sale.

[165] Scott Morton & Dinielli, supra note 15, at 29

[166] “Furthermore, the default makes the advertiser believe that search ads are very effective relative to display ads, so the advertiser has no reason to change the default.” Id.

[167] See About the Default MCF Attribution Models: Learn How Each MCF Model Assigns Conversion Credit, Google Analytics Help (last visited Nov. 1, 2022), Attribution Models, Google, https://support.google.com/analytics/answer/1665189?hl=en. For the Google Analytics 4 version of these attribution models currently being implemented, see [GA4] About Attribution and Attribution Modeling, Google Analytics Help (last visited Nov. 1, 2022), https://support.google.com/analytics/answer/10596866. Google even created a guide called “Beyond Last Click Attribution” to help advertisers select the most appropriate model. See Beyond Last Click Attribution: Official Guide to Attribution Modeling in Google Ads, Google Ads Help (last visited Nov. 1, 2022), https://support.google.com/google-ads/answer/7003286.

[168] See Joan Arensman & Wilfred Yeung, Move Beyond Last Click Attribution in AdWords, Google Blog (May 10, 2016), https://adwords.googleblog.com/2016/05/move-beyond-last-click-attribution.html.

[169] How Does Conversion Tracking Work?, Microsoft Advertising (n.d.) https://help.ads.microsoft.com/#apex/ads/en/56710/2.

[170] Nir Elharar, How to Choose the Right Marketing Attribution Model for Your Content, Outbrain (Apr. 8, 2019), https://www.outbrain.com/blog/marketing-attribution-model-content.

[171] Texas Complaint, supra note 1, at ¶ 195.

[172] Scott Morton & Dinielli, supra note 15, at 17

[173] Texas Complaint, supra note 1, at ¶ 351.

[174] See Demsetz, supra note 81.