On October 20, 2023, the Knight Institute will host a closed convening to explore the question of jawboning: informal government efforts to persuade, cajole, or strong-arm private platforms to change their content-moderation policies. Participants in that workshop have written short notes to outline their thinking on this complex topic, which the Knight Institute is publishing in the weeks leading up to the convening. This blog post is part of that series.


Jawboning needs to be understood in systemic, and not merely episodic, terms. Content moderation is a complex and dynamic system, not simply the aggregation of lots of individual decisions about individual posts. And so, the most difficult questions about government influence with content moderation are those about systemic effects, rather than individual ones. No one disputes (or no one should dispute) that if a government actor coerces a platform to take down a specific piece of content, that is a violation of the First Amendment. Much more complicated, though, is the question of what kind of systemic and ongoing interactions government actors can have with platforms, where the specific downstream effects are much harder to identify. How far can government actors go in informing platforms’ content moderation policies, if they never mention any particular post or user? What happens when a platform requests (rather than is pressured by) governmental input? Does it matter if governmental flags to platforms are often ignored by platforms, if they often result in takedowns?

These kinds of questions create fundamental challenges for case-by-case judicial adjudication of jawboning claims. As is characteristic of almost all First Amendment adjudication, jawboning cases tend to be highly particular in their focus. They focus on individual utterances and acts of government actors, and attempt to trace their impact on individual instances of speech.

Take Kennedy, Jr. v. Warren. The case concerned a letter written by Sen. Elizabeth Warren to Amazon’s CEO, raising concerns about the platform’s recommendation algorithms and their role in facilitating the spread of COVID-19 misinformation. The senator pointed to the plaintiffs’ book as promoted by Amazon in search results despite being “rife with misinformation” and perpetuating “dangerous conspiracies about COVID-19.” To determine if the senator crossed the line between lawful persuasion and unlawful coercion, the Ninth Circuit carefully parsed the words of the letter alone. In determining whether Warren possessed “regulatory authority” over Amazon, the court focused on the fact that “as a single Senator,” Warren “has no unilateral power to penalize Amazon for promoting” any book. And in evaluating whether Warren had threatened Amazon, the majority found that “Senator Warren never hinted that she would take specific action to investigate or prosecute Amazon” for promoting COVID-19 misinformation.

But reading this opinion feels like the Ninth Circuit has been living under a rock. Sen. Warren is not just any Senator—she is one of the nation’s most recognizable, once launching a high-profile presidential campaign pledging to break up Big Tech, and specifically Amazon. She is in the vanguard of a movement that now has serious momentum at the FTC. She has introduced bipartisan legislation to create a new commission to regulate online platforms, and written op-eds accusing them of “ruining lives.” She has publicly clashed with Amazon for years, and even traded snarky tweets. It may be true (as I myself tend to think) that none of this should take the senator’s letter into the realm of the unconstitutional, given the public nature of the letter and the value of government speech on matters of public concern. Nevertheless, these facts should at least be taken into account in the analysis. But for the Ninth Circuit, this broader context was beyond the purview of the particular dispute, which was about a particular letter and a particular book.

Or take O’Handley v. Weber. This case concerned the California Secretary of State’s office’s use of Twitter’s Partner Support Portal to flag tweets containing election misinformation to the platform. The court’s analysis again focused on the specific tweet flagged, and the specific communication by the Secretary of State’s office to Twitter, to find that the government “did nothing more than make a request with no strings attached.” But the court did not ask why the compliance rate with such requests was so high (98 percent), or what the surrounding circumstances of the relationship between the Secretary of State and Twitter were. Indeed, because the district court found nothing problematic in the individual communication, it granted a motion to dismiss without even the discovery that could have provided relevant context to the individual flag at issue.

This kind of analysis is what the adversarial justice system encourages but risks missing the jawboning forest for the trees. It asks: “Can we identify any particular statement threatening enough?” and “Did that statement relate to the particular piece of content in this case?” But it does not ask deeper questions about how government power is being wielded in the construction of the public sphere more broadly. This artificial focus on the individual facts before the court was always too narrow, but has become especially so in the age of these ongoing, extensive and often collaborative relationships that platforms have developed with government actors.

Enter, the Fifth Circuit Court of Appeals decision in Missouri v. Biden. This per curium decision, by contrast, zooms right out and purports to take a systemic approach. Much of the commentary on the opinion has rightly focused on its misleading account of both the facts and the law. But the court also adopted a fundamentally different approach to understanding jawboning and the influence of government actors on platforms. The court’s opinion tells a sweeping narrative of how government officials pressured social media platforms to address “misinformation” on their services, and suggests it is important to evaluate those relationships as a whole rather than by focusing on specific things government actors said or did and any specific pieces of content that the platforms removed as a result. It is striking that not once does the opinion explicitly trace a connection between a specific threat or demand issued by a government actor and a specific identified piece of content that a platform removed as a result, even though the district court opinion had done so. Rather than looking at individual statements of governmental officials, the court emphasized “the ‘tenor’ of the parties’ relationship and the conduct of the government in context.” The relevant governmental power that was being abused is left vague—the court saying that “a generalized, non-descript means to punish” the platforms could suffice.

Most striking of all is the “significant encouragement” branch of the court’s analysis. The court used the test of “coercion” in determining whether government officials crossed a constitutional line in their communications about removing specific posts and users from the platforms. But when asking whether government officials had violated the First Amendment by significantly encouraging platforms to moderate content, it focused on whether those government actors had so influenced the upstream moderation policiesof the platforms, that it was irrelevant whether they were involved in the downstream individual decisions. As the court said when discussing the actions of the White House officials and Surgeon General’s office:

By pushing changes to the platforms’ policies through their expansive relationship with and informal oversight over the platforms, the officials imparted a lasting influence on the platforms’ moderation decisions without the need for any further input. In doing so, the officials ensured that any moderation decisions were not made in accordance with independent judgments guided by independent standards. (emphasis added)

Or in the case of the FBI, the problem in the court’s eyes was that while platforms may have made their own decisions, “those decisions were made subject to commandeered moderation policies.” In relation to the CDC, “the platforms’ decisions were not made by independent standards, but instead were marred by modification from CDC officials.” This is a doctrinal innovation that attempts to move the level of analysis away from individual decisions and asks instead about the way governmental influence has been wielded overall. The Fifth Circuit opinion is clearly attempting to describe the system of jawboning that it thinks exists.

This is an important lens to bring to jawboning analysis. It has long been recognized that the incentives of internet platforms make them especially poor guardians of their users’ speech interests. A formalistic analysis that requires specific acts of coercion and compliance before a constitutional violation can be found will simply result in the same government power being exercised through more careful emails that do not trip any constitutional wires. And if this does not bother you in the context of COVID-19 misinformation, perhaps it might in the context of government pressure on platforms to take down speech it says is related to terrorism, or adult content, or even drill music.

While the systemic lens is important, the Fifth Circuit decision went too far in the opposite direction. The opinion zooms so far out that it talks about “the platforms” as if they are a monolith, and as if all of them had exactly the same interactions with government actors. The court barely acknowledged that governmental requests to platforms to remove content were frequently unsuccessful. It did not spare any consideration for the important First Amendment values served by government speech. And it was unsustainably vague in the guidance it gave to government actors about where the line is between unproblematic regular meetings, trading of information, or advocating for certain policies, and the “something more” that the court found was present in this case.

Even so, the Fifth Circuit’s decision holds important lessons. First, the divergence between the outcomes in the Ninth Circuit and the Fifth Circuit show the impact that the level of analysis can have on outcomes and forces the question of the right level of generality at which to assess jawboning claims. We should not throw the Fifth Circuit’s systemic approach out with its motivated reasoning bathwater. The opinion is right to take a broader lens; but courts should be more evidence-based when they do. What these cases point to is the need for a middle ground approach that is not too focused on individual instances but not so abstracted that there is no attention to relevant details. Only then can all the equities jawboning claims implicate be properly accounted for.

Second, a systemic approach also forces us to ask questions about the point at which First Amendment protections should intervene. The First Amendment is a system of prophylactic rules, and jawboning is one such problem that is likely to be better addressed by upstream rules to reduce the danger of downstream unconstitutional speech suppression. Structural mandates that mitigate the potential for bias or irrelevant considerations infecting the decisionmaking process, carefully crafted transparency into governmental communications with platforms, and institutional audits and compliance with rules of engagement for such communications, are much more likely to be effective at addressing the harm that unconstitutional jawboning does to the public sphere than a focus on case-by-case redress for individual plaintiffs.

A systemic problem requires a systemic analysis and response. This is the frame that courts, yes, but also scholars and legislatures should be bringing to bear to the questions jawboning raises, so that we do not miss either the jawboning forests or the trees.