The Platform Bargain Is Broken

In 2010, Jack Dorsey gave an interview in which he described Twitter as a neutral utility — a town square where ideas competed on their merits, where the best arguments won, where democratic discourse flourished free from the editorial interference of traditional media gatekeepers. The metaphor was genuinely appealing to a generation that had grown up distrusting mainstream media and finding community in digital spaces. The town square felt democratic in a way that the New York Times editorial board emphatically did not.

The town square metaphor was also, in retrospect, a piece of self-serving public relations that obscured the actual structure of what Twitter — and Facebook, and YouTube — were building.

A town square has no algorithm. It does not select which speakers the crowd hears based on what will maximize how long the crowd stays. It does not sell the crowd’s attention to advertisers. It does not maintain a detailed psychological profile of each person in the crowd and use that profile to determine which speakers they encounter. Twitter, Facebook, and YouTube were not town squares. They were attention-harvesting businesses that found it useful to describe themselves as town squares during the period when regulatory scrutiny was still forming.

What the Bargain Actually Said

The implicit bargain between social media platforms and democratic society was never explicitly stated. Had it been stated, it would have looked something like this:

We, the platforms, will host public political discourse, including political advertising, and will profit enormously from doing so. In exchange, we will take reasonable steps to ensure that political discourse on our platforms does not systematically degrade the quality of democratic decision-making. We will remove content that is demonstrably false and harmful. We will identify and suppress coordinated inauthentic behavior. We will ensure that political advertising is disclosed and attributable. We will prioritize the integrity of elections over other considerations during election periods.

Whether any platform ever genuinely intended to keep this bargain is debatable. What is not debatable is that the business models of the platforms have always been in structural conflict with keeping it.

Related reading

The conflict: the content that most effectively engages users — that maximizes the time-on-platform metric that drives advertising revenue — is not the content that best serves democratic deliberation. Outrage drives engagement. Fear drives engagement. Tribal confirmation drives engagement. The nuanced policy analysis, the candidate comparison based on documented positions, the careful evaluation of competing claims — these are hard to optimize for engagement because they require cognitive work that reduces the pleasure response that keeps people on the platform.

A platform that genuinely prioritized democratic information quality would perform worse on the engagement metrics that determine its stock price. This is not a hypothetical conflict. It is the actual operating environment of every major social media company.

The 2026 Policy Announcements: Reading the Fine Print

In early 2026, Meta, Google/YouTube, and TikTok all published updated policies for AI-generated political content in the election year. The announcements were prominently covered. The specific policy details received less attention.

Meta’s policy for AI-generated political content requires advertisers to disclose “material” use of AI in political advertising. The definition of “material” use is determined by Meta’s internal review process, not by any external standard. Disclosure takes the form of a small label visible on the advertisement. The policy applies to paid advertising; it does not apply to organic content distributed without payment. It does not apply to AI-generated content distributed through networks of fake personas, because that content is not “advertising” in Meta’s policy framework.

Google’s policy prohibits AI-generated political advertising that uses “synthetic imagery of real people in a false or misleading context.” The enforcement mechanism is a combination of automated detection and human review. Google does not publish its detection accuracy rates or its false negative rates (the percentage of prohibited content that passes through undetected). Independent researchers who have tested the system estimate that a significant fraction of content that would technically violate the policy does pass through, particularly content in non-English languages and content that has been lightly modified from purely AI-generated originals.

TikTok’s policy prohibits AI-generated “deepfakes” in political content and requires labeling of AI-generated content in political advertising. TikTok does not have an established track record of election integrity enforcement comparable to Meta and Google, and its ownership structure — ByteDance, a Chinese company — creates a structural conflict of interest in enforcing against Chinese-origin influence operations that most commentators decline to name directly.

The pattern across all three policy frameworks: specific rules that can be pointed to as evidence of good faith, enforcement mechanisms that are insufficiently resourced and transparently inadequate to the scale of the problem, and definitional boundaries that exclude the categories of AI political content that are most harmful.

The Revenue Problem Does Not Resolve

The structural argument — that political advertising revenue creates a conflict of interest that prevents genuine platform enforcement of political content quality — has been made before, and the platforms have responded by pointing to their policies, their investments in trust and safety teams, and their willingness to block some political advertising categories.

These responses do not address the structural argument.

Google earned approximately $1.2 billion in political advertising in the 2024 cycle. Meta earned approximately $900 million. These are not trivial numbers, but they are also not the majority of either company’s revenue. The companies’ reluctance to aggressively police political content cannot be explained purely by protecting political advertising revenue — they could afford to lose it.

The better explanation is the dual role of political content: political advertising is revenue, but political content — the organic posts, the news articles, the political arguments — is engagement fuel that makes the platform valuable for all advertising, not just political advertising. A platform that rigorously polices political content and produces a calmer, less emotionally activating information environment would likely see reduced overall engagement and reduced overall advertising revenue across all categories. The political content isn’t just valuable as advertising inventory. It is the engagement driver that makes the platform’s total audience valuable.

This is the deeper structural conflict that the platforms’ policy announcements do not address and cannot address without changing their business models. The problem is not political advertising revenue. The problem is that inflammatory political content is, for the engagement-based advertising business, an asset rather than a liability.

Related reading

What Independent Research Shows About Platform Enforcement

The most rigorous external evaluation of platform political content enforcement comes from the Stanford Internet Observatory, the MIT Media Lab’s Center for Constructive Communication, and similar academic research groups that systematically test platform policies against real-world content.

The consistent finding, across multiple research groups and multiple election cycles, is that platform enforcement of political content policies is significantly more effective against content from small accounts and foreign actors than against content from large domestic accounts with significant following and engagement history. This is partly a prioritization effect — large accounts get more human review time and more rapid appeals resolution — and partly an enforcement constraint — removing content from a large, high-engagement account produces significant user backlash and press attention that small-account removal does not.

The practical implication: platform enforcement has a thumb on the scale that favors political actors with existing large followings and penalizes smaller actors. When AI-generated disinformation is distributed through established accounts with large audiences — a tactic increasingly favored by sophisticated domestic political operations that buy access to established accounts — it receives less effective enforcement than equivalent content from small, foreign-origin accounts.

A second consistent finding: enforcement is faster and more accurate for content in English than for content in other languages. The platforms’ trust and safety teams are disproportionately English-speaking, the automated detection systems are primarily trained on English-language content, and the human review capacity for non-English languages is thinner. Filipino voters, Brazilian voters, Pakistani voters, and Indian voters in regional languages are significantly less protected by platform enforcement than American English-speaking voters.

This geographic and linguistic bias in enforcement is not discussed in the platforms’ public communications, because discussing it would require acknowledging that platform protections are unequally distributed in ways that correlate with market value.

The European Experiment

The European Union has taken a more aggressive regulatory approach to platform political content than any other major democratic jurisdiction, and the results are instructive about what regulation can and cannot achieve.

The Digital Services Act, in force since 2024 for the largest platforms, requires designated “very large online platforms” (those with over 45 million monthly EU users) to assess and mitigate systemic risks to democratic processes, including from AI-generated political content. Platforms must report these risk assessments to the European Commission and accept audits from approved independent researchers.

The practical consequences have been significant in some respects. Meta, Google, and TikTok have all increased their investment in trust and safety operations covering EU elections in response to DSA compliance obligations. The German federal election in February 2026 saw more aggressive platform enforcement than comparable elections in non-EU countries — though, as noted, this did not prevent a significant foreign influence operation from running for months.

The DSA’s limitations are equally instructive. Enforcement is slow — the Commission’s investigation processes take months to complete during which non-compliant behavior continues. Fines, while nominally up to 6 percent of global annual revenue, have been smaller in practice. The DSA’s risk assessment requirements place the compliance burden on platforms whose incentive is to demonstrate compliance rather than achieve it.

Most significantly: the DSA addresses platform obligations around political content. It cannot address the underlying structural conflict between engagement-maximizing business models and democratic information quality. A platform can comply fully with DSA requirements while still operating an algorithm that systematically amplifies inflammatory political content, because the DSA regulates specific practices rather than the business model that generates those practices.

The Business Model Question Nobody Asks

The only intervention that directly addresses the structural conflict between platform business models and democratic information quality is changing the business model.

An attention-based advertising model, in which user engagement drives advertising inventory value, creates strong incentives to maximize emotional arousal, which creates strong incentives to amplify inflammatory political content, which creates strong incentives against rigorous political content enforcement. This is not an accident or a corruption. It is the logical output of the incentive structure.

Related reading

A subscription-based model, in which platform revenue is decoupled from advertising and therefore decoupled from engagement maximization, would remove this structural incentive. A platform that earns revenue from subscriber fees rather than advertising has reduced financial incentive to maximize engagement at the expense of content quality. The incentive structure changes.

Why don’t platforms change their business models? Because the advertising model generates vastly more revenue than subscription models do at the relevant scale. Meta’s advertising revenue in 2025 was approximately $165 billion. A subscription model generating equivalent revenue would require charging every one of Meta’s approximately three billion monthly users around $55 per year — which competitive pressure would make nearly impossible to achieve.

The business model argument is therefore also an argument about market structure. A platform industry with genuine competition — in which users could easily leave for platforms with better information quality — might produce competitive pressure toward information quality. The current industry, characterized by strong network effects that make the dominant platforms near-monopolistic for their core use cases, does not produce this pressure.

Structural reform of the platform industry — through antitrust action, through interoperability requirements that reduce network effect lock-in, through public investment in alternative platform infrastructure — is the only intervention that addresses the business model conflict directly. It is also politically the most difficult, most delayed, and most contested.

The broken bargain between platforms and democracy is not an accident of bad faith or insufficient effort. It is the inevitable output of a specific economic structure. Fixing it requires changing that structure, and the political will to change it has not yet materialized.