EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Truth Markets and the Consensus Trap

Timothy Graham

M/C Journal · 21.04.2026 · 29 (2)

DOI: 10.5204/mcj.3250 ↗

Kurz gefasst

Fachartikel in der Zeitschrift M/C Journal, online erschienen am 21.04.2026. Verfasst von Timothy Graham. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Epistemic Infrastructure At 13:07 on 23 April 2013, the Associated Press (AP) tweeted: “Breaking: Two Explosions in the White House and Barack Obama is Injured”. Within seconds, automated trading systems detected the news. The Dow Jones dropped 143 points. The S&P (Standard & Poor’s) 500 lost $136.5 billion in market value. Six minutes later, everything returned to normal. The tweet was fake – the AP account had been hacked by the Syrian Electronic Army. The hack was a security failure. But the market’s reaction revealed something larger: by 2013, Twitter had become the world’s epistemic infrastructure for breaking news – the system through which claims about current events circulated, were attributed to authoritative sources, and thereby gained credibility. Markets monitored it. Governments issued official statements through it. Journalists treated it as a primary source. The market panicked not because of a blue checkmark but because the tweet appeared to come from the Associated Press – one of the most trusted news organisations on Earth. The verification badge confirmed identity – this really is AP – but on infrastructure this central, identity authentication became inseparable from epistemic authority. To be verified on Twitter was to have your claims amplified through the world’s most consequential channel for breaking news. This article examines what happened when a platform that positioned users in a reportorial posture became a global infrastructure through which all manner of claims flowed. Twitter’s central prompt – originally “What are you doing?”, changed to “What’s Happening?” in November 2009 (Burgess and Baym) – oriented users as witnesses to reality, whether the reality of personal experience or of public events. The prompt presupposed a relationship between speaker and world: you have encountered something, experienced something, and you are reporting it. This framing enacted what John Durham Peters identifies as the discursive structure of witnessing – the triangulation of presence, testimony, and authority that grounds a speaker’s claim in their relationship to the real (Peters). Yet as Zizi Papacharissi has shown, platforms like Twitter generate affective publics in which personal sentiment and public discourse flow through the same channels; “What’s Happening?” invited both registers, but always within an implicit orientation toward correspondence between utterance and world (Papacharissi). Empirical assertions, political interpretations, shitposts, contested narratives, normative arguments: they all moved through the same system, governed by the same design features, sorted by the same engagement metrics. The problems that followed stem not from a betrayal of the platform’s stated purpose, but from a mismatch between the epistemic posture the infrastructure assumed or imagined and the epistemic plurality of the discourse it came to carry. A note on terminology: I use “epistemic infrastructure” to denote the sociotechnical systems through which claims circulate, credibility is conferred, and validation occurs. This follows Susan Leigh Star’s understanding of infrastructure as the often-invisible arrangements that enable and constrain knowledge production. I reserve “truth” for the property of propositions – what makes a claim true – and distinguish this from “truth-determination”, the epistemic processes through which we establish whether claims possess that property. This distinction matters because, as I will argue, the failures of platforms like Twitter are primarily epistemic failures – failures of verification procedures – rather than failures concerning truth itself. Truth Markets and Truth-Determination In 2015, communication scholar Jayson Harsin introduced a concept that illuminates what was happening. Drawing on Foucault’s notion of “regimes of truth” – the mechanisms through which societies “distinguish true and false statements” and designate “those who are charged with saying what counts as true” (130) – Harsin argued we had shifted toward “regimes of posttruth” (ROPT), characterised by proliferating “truth markets” (327). Harsin’s concept requires careful interpretation. “Truth markets” are not markets in truth itself – as if truth were a commodity that could be bought and sold. Rather, they are competitive arenas of truth-determination: fragmented spaces where claims compete for credibility without authoritative institutions capable of settling disputes. In traditional regimes of truth, authoritative institutions – credentialed journalism, scientific consensus, official records – coordinated society’s processes for validating claims. In truth markets, no such coordination obtains. Every claim competes for attention, and engagement metrics determine visibility regardless of whether claims are true. Crucially, Harsin observed, “there is no authoritative Debunker” in truth markets. Fact-checkers and verification organisations proliferate, yet none “can suture fragmentation through a judgment for an entire society, and certainly cannot endure temporally” (331). A correction might reach thousands while the original claim reaches millions. Even when corrections spread widely, the false claim continues circulating long after attention has moved elsewhere. This temporal asymmetry – the structural advantage of first claims over corrections – would prove central to understanding Twitter’s trajectory as epistemic infrastructure. Harsin’s framework also highlights how truth markets demand participation. Unlike older regimes where citizens adhered to institutionally sanctioned knowledge, ROPT require “popular attention to/participation in its discursive games” (331). Resource-rich actors exploit this participatory dynamic, multiplying claims “whose meaning, if not veracity, is not easily or quickly confirmed” (331). The goal is not always persuasion but attention management – occupying the field of perception to shape what claims gain visibility. Twitter became the infrastructure through which these truth markets operated. The platform’s design – including the reportorial posture embedded in “What’s Happening?” – shaped market dynamics: which claims spread, which languished, whose voice carried weight. That posture posed fewer problems when Twitter primarily carried claims about current events – the kind of information the AP hack exploited. “Explosions at the White House” is an empirical claim whose truth is grounded in correspondence to observable states of affairs. Within minutes, other sources can verify whether explosions occurred. The truth market corrects relatively quickly because participants share epistemic standards: everyone agrees that the claim is true if and only if it corresponds to what actually happened. The problem emerged when Twitter became the channel for claims that cannot be verified the same way – political interpretations, normative arguments, claims whose truth, if they possess it, is grounded not in simple correspondence to observable facts but in coherence with frameworks, values, or standards that communities do not share. The same infrastructure, the same design features, the same engagement-driven sorting – but fundamentally different epistemic domains requiring different verification approaches. “The Free Speech Wing of the Free Speech Party” Twitter’s leadership never recognised that they had built epistemic infrastructure. In 2012, Tony Wang, Twitter’s UK general manager, described the company as “the free speech wing of the free speech party”. CEO Dick Costolo repeated this framing. The philosophy was clear: Twitter was a neutral conduit. Information flowed through it; the company did not adjudicate that information. This was a category error. Neutral conduits do not have verification systems that determine whose voice carries institutional authority. They do not have algorithms that decide which claims get amplified. They do not have trending topics that shape what millions perceive as significant. Twitter had all of these design features, and each one shaped the epistemic dynamics operating within it. The “free speech wing” philosophy meant Twitter disclaimed responsibility for design choices that determined which claims would gain credibility and which would fade into obscurity. The consequences were visible early. During the Boston Marathon bombing in April 2013 – the same month as the AP hack – researcher Kate Starbird found that 29% of the most viral content consisted of rumours and false claims (Starbird et al.). Corrections emerged quickly, often within the same hour. But they did not spread as far or persist as long. For claims about current events, this was damaging but self-correcting: eventually the verified account of what happened in Boston prevailed because everyone applied shared epistemic standards for evaluating empirical claims. Throughout this period, Twitter had no policy addressing false claims. None. While Facebook introduced “disputed” labels in 2016, Twitter would not implement any such labelling until 2020 – a four-year gap. For empirical claims about current events, this might have been defensible; verification eventually occurs through independent sources. But Twitter was no longer only carrying such claims. It had become infrastructure for political discourse, ideological argument, and contested interpretations where no single verification procedure could establish which claims were true. In July 2016, Twitter opened verification to public applications. The move exposed confusion about what verification meant on infrastructure this consequential. White supremacist Jason Kessler, organiser of the 2017 Charlottesville rally, was verified. So was white nationalist Richard Spencer. Twitter had confirmed their identities – verification performing its technical function. But on infrastructure that had become the global channel for authoritative claim-making, verification appeared to confer epistemic legitimacy. By November 2017, Twitter reversed course, revoking verification for “promoting hate” and pausing the program entirely. The reversal revealed what Twitter had never articulated: on epistemic infrastructure this central, identity authentication and epistemic authority cannot be cleanly separated. The Exploitation State-sponsored influence campaigns did not create the problems with Twitter’s epistemic infrastructure – they exploited them. Russia’s Internet Research Agency (IRA) understood what Twitter’s leadership had failed to grasp: that infrastructure built around a reportorial posture was now carrying contested political claims, and the same design features that spread breaking news would spread divisive content just as effectively. The infrastructure optimised for engagement, not truth. Twitter’s initial response revealed institutional blindness. In January 2017, the company reported finding just 22 accounts meeting its criteria for coordinated influence. U.S. Senator Mark Warner told Twitter it had demonstrated “an enormous lack of understanding ... of how serious this issue is”. By October 2018, Twitter had identified 3,841 IRA accounts and released archives of nearly 10 million tweets dating back to 2009 (Twitter, “Enabling Further Research”). The scale of what had been missed was staggering. The U.S. Senate Select Committee on Intelligence found these accounts had “engaged with several distinct communities ... which exhibited only minimal overlap” (vol. 2). The IRA had segmented the American public into separate epistemic communities – conservatives, progressives, Black communities – and targeted each with content designed to inflame existing divisions. Research on similar campaigns demonstrates how coordinated networks exploit platform architectures to manufacture apparent grassroots support. Analysis of the #IStandWithPutin campaign following Russia’s invasion of Ukraine found that bot networks could boost hashtags into trending lists, triggering what Phillips calls the “oxygen of amplification” – where even critical engagement inadvertently extends a campaign’s reach (Graham, “How Propaganda”). These campaigns succeeded not by circumventing Twitter’s infrastructure but by mastering its logic: engagement determines visibility, regardless of whether claims are true or even sincere. Twitter co-founder Jack Dorsey’s September 2018 congressional testimony finally acknowledged the problem: “we aren’t proud of how that free and open exchange has been weaponized … that’s not a healthy public square”. The “free speech wing” philosophy was over. But the epistemic infrastructure remained unchanged in its fundamental architecture. The Monetisation Elon Musk’s 2022 acquisition transformed the infrastructure’s political economy. When Twitter Blue launched at $8/month, verification – the mechanism that had filtered whose claims carried institutional authority – became purchasable. A fake account impersonating pharmaceutical company Eli Lilly tweeted: “we are excited to announce insulin is free now”. The company’s stock dropped. The system that once distinguished institutional voices from anonymous accounts was now available to anyone willing to pay. Ad revenue sharing, introduced in July 2023, created direct financial incentives for engagement regardless of accuracy. NewsGuard found that 74% of the most-engaged posts spreading Israel-Hamas misinformation came from verified paying accounts. The epistemic infrastructure that had become central to global information now paid users to produce engaging content – and engagement metrics reward outrage over nuance, sensationalism over careful qualification. This confirms what Harsin’s framework predicts: truth markets require “popular attention to/participation in its discursive games” (331). Monetisation financially incentivises that participation without regard to whether claims are true. Meanwhile, research documents how the platform’s bot problem has worsened considerably following the acquisition, with AI-powered accounts flooding the platform with low-quality content that mimics authentic discourse (Graham and FitzGerald). Grok, launched as a “maximum truth-seeking AI”, generated incorrect citations 94% of the time (Columbia Tow Center) and spread false claims about U.S. ballot deadlines that five Secretaries of State publicly corrected (Simon et al.). The platform owner was now operating as an automated truth-arbiter through a system that routinely failed at basic verification. The Consensus Trap Community Notes, rebranded from Birdwatch in November 2022, represents Twitter’s most sophisticated attempt to govern the truth markets operating on its infrastructure. The system is genuinely innovative. Users write notes contextualising potentially misleading content. A “bridging algorithm” surfaces notes that receive approval from users who typically disagree (Wojcik et al.). Unlike traditional fact-checking, which critics dismiss as biased, Community Notes derives legitimacy from cross-partisan consensus. Other platforms have begun adopting similar approaches. Where it works, it works fairly well. Research shows 97.5% accuracy on COVID-19 vaccine notes (Ayers et al.) and 46% reduction in sharing when notes appear (Slaughter et al.). For scams, obvious hoaxes, and verifiable factual errors, the system provides real epistemic value. These are claims whose truth is grounded in correspondence to verifiable states of affairs – the kind of claims that align with the reportorial posture embedded in the platform’s design. “This image is from 2019, not 2024”. Anyone can verify the date. Consensus is achievable because participants share standards for evaluating whether the claim corresponds to reality. But Community Notes reveals the consensus trap: a term adapted from Daniel Fiorino’s analysis of regulatory governance, where agencies commit to consensus-based decision processes that participants can never achieve (Fiorino). Applied to platform governance, the consensus trap describes how the system succeeds precisely in domains where independent verification is already possible, and fails precisely where epistemic assistance is most needed. For contested political claims – where communities disagree not merely about facts but about the frameworks through which claims should be evaluated – consensus is structurally unattainable. The bridging algorithm requires agreement from users who typically disagree. On partisan content, they cannot agree. The note never appears. Analysis by the Digital Democracy Institute of the Americas found that over 90% of proposed Community Notes remain unpublished – they never achieve the consensus threshold required for display (see also Fundación Maldita.es for a comprehensive empirical analysis). Political content rarely crosses that threshold. Meanwhile, viral misinformation peaks within four hours of posting; notes average eighteen hours or more to appear. Research shows notes appearing after 48 hours have “almost no effect” on spread (Arjmandi-Lari et al.). For contested claims – the claims that now dominate Twitter – correction either never arrives or arrives far too late. This confirms Harsin’s prediction that in truth markets, no mechanism “can suture fragmentation through a judgment for an entire society, and certainly cannot endure temporally” (331). The consensus trap is not a bug in Community Notes; it is intrinsic to its design. The system achieves legitimacy precisely by requiring cross-partisan agreement – that requirement is what makes it trustworthy. But that same requirement renders it ineffective for contested content. Community Notes works for claims that align with the reportorial posture embedded in the platform’s design. Each new post prompts the user with “What’s Happening?” – an invitation to witness, to testify to reality, presuming a correspondence between utterance and world. Most posts do not adhere to this logic of correspondence. Yet “What’s Happening?” remains the platform’s persistent, silent assumption that what follows will bear a reportorial relationship to reality. Community Notes succeeds precisely within this assumed posture, and fails for the epistemic infrastructure Twitter actually became: a global platform carrying claims across every domain of discourse. Epistemic Pluralism and Platform Design The philosopher Michael Lynch has argued for what he calls “alethic pluralism” – the view that truth, while a single unified property, can be manifested or grounded differently across different domains of discourse (Lynch). This is not relativism – Lynch does not claim there are “multiple truths” – but rather that what makes a claim true – the underlying property that grounds its truth – varies by domain. For empirical claims about observable states of affairs, truth is grounded in correspondence: the claim is true if it accurately represents how things are. For claims in other domains – mathematics, ethics, perhaps politics – truth may be grounded differently, in coherence with broader systems or in meeting standards of warranted assertibility. Lynch’s framework has epistemic implications that bear directly on platform governance. If truth is grounded differently across domains, then the procedures through which we verify claims – the epistemic methods by which we determine whether claims possess the property of truth – must also vary. Procedures adequate for verifying correspondence (checking claims against observable facts) may be inadequate for domains where truth is grounded otherwise. This illuminates the consensus trap. Community Notes is an epistemic mechanism well-suited to claims where truth is grounded in correspondence. For such claims, verification is straightforward: we check whether the claim matches verifiable reality. “The image is from 2019” can be verified by anyone with access to reverse image search. Consensus is achievable because the epistemic procedure is shared – everyone agrees on how to determine whether the claim is true, even if they initially disagree about whether it is. For claims where truth is grounded differently – in coherence with normative frameworks, political values, or interpretive schemas that communities do not share – Community Notes’ consensus mechanism cannot function. Not because such claims lack truth values, but because participants lack shared procedures for verification. They disagree not merely about whether a claim is true but about how to determine whether it is true. Robert Fogelin termed such cases “deep disagreements” – disagreements that reach beyond specific conclusions to the fundamental epistemic frameworks through which evidence is evaluated, where the ordinary mechanisms of rational adjudication lose their grip (Fogelin). The bridging algorithm requires agreement from people who disagree; but when disagreement is deep in Fogelin’s sense – extending to epistemic standards themselves – no bridge is possible. The lesson for platform design is not that contested claims should be left ungoverned – that returns us to the engagement-driven chaos of unmoderated truth markets. The lesson is that epistemic infrastructure this consequential requires plural verification mechanisms matched to the domains of discourse it carries. Consensus-based correction may work for verifiable empirical claims. Where disagreements are deep, different approaches are needed: transparency about the existence of disagreement, clear labelling of claims as contested, or mechanisms that surface the structure of disagreement rather than attempting to resolve it. The goal is not to eliminate disagreement – that is neither achievable nor desirable. The goal is to ensure that straightforwardly false empirical claims can be corrected even when broader disputes remain unresolved, and that the epistemic infrastructure does not systematically advantage false claims through engagement dynamics. This is the design principle that Harsin’s framework suggests: since no single mechanism “can definitively debunk truths by suturing multiple audience/market/network segments” (332), infrastructure must be epistemically plural – matching verification mechanisms to the domains of claim-making at stake. Conclusion The AP hack that opened this article demonstrated that Twitter had become global epistemic infrastructure for breaking news. The two decades that followed demonstrated what happens when infrastructure built around a reportorial posture – positioning users as witnesses to reality – becomes the channel through which all claims flow, shaped by the same design features and sorted by the same engagement metrics that privilege the sensational over the accurate. Twitter’s designers made choices about verification, amplification, and monetisation without understanding their epistemic consequences. The infrastructure they built now shapes how billions of people encounter claims about the world. What was designed can be redesigned. But that requires seeing clearly what the current architecture actually does: it treats claims from all domains as equivalent, sorts them by engagement rather than epistemic merit, and through the consensus trap fails systematically at the contested claims that matter most. The question for the next twenty years is whether we can build epistemic infrastructure adequate to the plurality of claims it must carry – infrastructure that neither imposes false consensus nor abandons verification entirely, but matches epistemic mechanisms to epistemic domains. That is the architectural challenge we have inherited. References Arjmandi-Lari, Zahra, et al. “Threats to the Sustainability of Community Notes on X.” arXiv, Oct. 2025: 2510.00650. Columbia Tow Center. “AI Search Engines Fail to Produce Accurate Citations.” Nieman Lab, 10 Mar. 2025. Ayers, John W., et al. “Characteristics of Community Notes Addressing COVID-19 Vaccine Misinformation on X (Formerly Twitter).” JAMA 331.19 (2024): 1670–72. Burgess, Jean, and Nancy K. Baym. Twitter: A Biography. New York: NYU P, 2020. Digital Democracy Institute of the Americas. “A Deep Dive into X’s Community Notes: An Analysis of English and Spanish Contributions Between 2021 and 2025.” July 2025. Dorsey, Jack. Testimony before the U.S. Senate Select Committee on Intelligence. 5 Sep. 2018. Fiorino, Daniel J. “Regulatory Policy and the Consensus Trap: An Agency Perspective.” Analyse & Kritik 19.1 (1997): 64–76. Foucault, Michel. “Truth and Power.” Power: Essential Works of Foucault, 1954–1984. Ed. James D. Faubion. Vol. 3. New York: New Press, 2000. 111–33. Fogelin, Robert J. “The Logic of Deep Disagreements.” Informal Logic 7.1 (1985): 1–8. Fundación Maldita.es. “Faster, Trusted, and More Useful: The Impact of Fact-Checkers in X’s Community Notes.” Maldita.es, Feb. 2025. Graham, Timothy. “How Propaganda Exploits the Infrastructure of Truth: A Case Study of #IStandWithPutin.” Critical Studies in Media Communication 42.1 (2025): 75–82. Graham, Timothy, and Kate M. FitzGerald. “Bots, Fake News and Election Conspiracies: Disinformation during the Republican Primary Debate and the Trump Interview.” Digital Media Research Centre, Queensland University of Technology, 2023. Harsin, Jayson. “Regimes of Posttruth, Postpolitics, and Attention Economies.” Communication, Culture & Critique 8.2 (2015): 327–33. Lynch, Michael P. Truth as One and Many. Oxford: Oxford UP, 2009. NewsGuard. “Blue-Checked, ‘Verified’ Users on X Produce 74 Percent of the Platform’s Most Viral False or Unsubstantiated Claims Relating to the Israel-Hamas War.” Oct. 2023. Papacharissi, Zizi. Affective Publics: Sentiment, Technology, and Politics. Oxford: Oxford UP, 2015. Peters, John Durham. “Witnessing.” Media, Culture & Society 23.6 (2001): 707–24. Phillips, Whitney. “The Oxygen of Amplification: Better Practices for Reporting on Extremists, Antagonists, and Manipulators Online.” Data & Society, 2018. Simon, Steve, et al. Letter to Elon Musk. 5 Aug. 2024. Slaughter, Isaac, et al. “Community Notes Produce Unreliable Misinformation Annotations on Social Media.” Proceedings of the National Academy of Sciences 122.38 (2025): e2417208122. Star, Susan Leigh. “The Ethnography of Infrastructure.” American Behavioral Scientist 43.3 (1999): 377–91. Starbird, Kate, et al. “Rumors, False Flags, and Digital Vigilantes: Misinformation on Twitter after the 2013 Boston Marathon Bombing.” iConference 2014 Proceedings, 2014. Twitter. “Enabling Further Research of Information Operations on Twitter.” Twitter Blog, Oct. 2018. U.S. Senate Select Committee on Intelligence. Report on Russian Active Measures Campaigns and Interference in the 2016 U.S. Election. Vol. 2. 2019. Wojcik, Stefan, et al. “Birdwatch: Crowd Wisdom and Bridging Algorithms Can Inform Understanding and Reduce the Spread of Misinformation.” arXiv, 2022: 2210.15723.

Zitieren

APA

Graham, T. (2026). Truth Markets and the Consensus Trap. M/C Journal, 29(2). https://doi.org/10.5204/mcj.3250

Vancouver

Graham T. Truth Markets and the Consensus Trap. M/C Journal. 2026;29(2). doi:10.5204/mcj.3250

BibTeX
@article{Graham2026,
  author = {Graham, Timothy},
  title = {Truth Markets and the Consensus Trap},
  journal = {M/C Journal},
  year = {2026},
  volume = {29},
  number = {2},
  doi = {10.5204/mcj.3250}
}
Kostenlos starten →