The empirical baseline: concentration, not saturation
The unifying empirical finding across this literature is that exposure to misinformation and low-credibility content is rare in the aggregate and heavily concentrated among a small minority of users. Budak2024-ef articulates this as a general thesis — public discourse has drifted from the evidence on prevalence, algorithmic causation, and societal harm — and the rest of the corpus supplies the data points. Bergeron-Boutin2026-yh and the related Appel2026-qr/noauthor_undated-bm studies, using Meta’s platform-scale data from the US 2020 election, show that untrustworthy-source and deceptive-network content reached only a small share of feeds on average, that a handful of networks and a tiny fraction of resharers drove the overwhelming majority of exposure, and that associations with downstream political outcomes evaporate once pre-exposure characteristics are controlled for. Lyons2026-ca extends this logic to health information, finding that low-credibility content is likewise rare overall but concentrated among older, more conspiracist-minded users, and travels mainly through referrals from other dubious sites rather than search or social media — a finding that complicates simple “platform algorithm” stories. Gonzalez-Bailon2024-rq and Rossi2023-847d5a9f extend the concentration finding geographically and structurally: most posts reach small audiences, only a sliver go viral, and even where the share of untrustworthy URLs rises (e.g., in German and Italian election years), the share of views attributable to them stays comparatively flat. Mosleh2024-op shows this concentration pattern is not a Twitter/Facebook artifact but recurs, with platform-specific variation, across seven platforms.
Complementing these exposure studies are causal and quasi-causal tests of whether such exposure, even where it occurs, actually moves attitudes or behavior. Allcott2025-jb finds that removing political ads from feeds for six weeks before the 2020 election had no detectable effect on knowledge, polarization, or perceptions. Brady2026-ln shows that even substantial algorithmic amplification of moralized/emotional content changes perceived social norms more than it changes people’s own behavior, and that toxic engagement itself is extremely concentrated (Gini = 0.93) — echoing the exposure-concentration finding at the level of production rather than consumption, a pattern Oswald2025-km generalizes as the “production-consumption gap”: a small vocal minority generates the visible discourse that both citizens and researchers mistake for public opinion.
Network- and node-level challenges to “coordination” narratives
A parallel strand recalibrates claims about who drives misinformation diffusion. Di-Marco2025-aa formally models the theoretical maximum influence coordinated accounts could achieve in retweet cascades and finds real UK-election coordinated accounts fall far short of it, with placement patterns statistically indistinguishable from random — suggesting coordinated inauthentic behavior is less pivotal than assumed, largely for want of resources and strategy. Pante2025-pq applies similar rigor to claims of inter-state collaboration in influence operations, arguing that without modern coordination detection and proper control datasets, prior claims of coordination don’t hold up. Together with Appel2026-qr’s finding that most reach from even the largest deceptive networks came from ordinary users resharing content rather than the networks’ own accounts, this cluster relocates agency away from shadowy coordinated actors and toward mundane, diffuse user behavior.
Measurement artifacts: how the field manufactures its own alarm
Several papers argue that part of the apparent scale of the misinformation problem is a methodological artifact of how researchers operationalize it. Giglietto2022-b30e8b4e shows that Meta’s 100-public-shares anonymization threshold, combined with shifting News Feed ranking, produces near-identical monthly trend lines across 46 wildly different countries — a platform-wide statistical artifact easily mistaken for a substantive finding (including a spurious “COVID misinformation surge” that actually predates the pandemic). Luhring2025-od audits NewsGuard, the workhorse source-trustworthiness database behind much of this literature, finding that continuous scores are fairly stable since 2022 but that binary trustworthy/untrustworthy cutoffs are exquisitely sensitive to single large outlets crossing the threshold, meaningfully distorting reported country-level “misinformation shares.” Hartmann2025-px’s systematic review of 129 echo-chamber studies makes an analogous point at the conceptual level: much of the field’s disagreement about whether echo chambers exist stems from incompatible operationalizations rather than incompatible realities. Goel2025-iq flips the source-based paradigm on its head, showing that domain-list measures miss a large phenomenon in which factually accurate mainstream articles are selectively co-shared to support misleading narratives — meaning source-level “misinformation” metrics may simultaneously overstate exposure to fringe content and understate the narrative misuse of legitimate journalism. Nenno2025-xa extends this critique cross-nationally, finding that news-value differences between flagged and unflagged content are statistically real but substantively small, and that WEIRD-derived typologies of “what makes something newsy” travel poorly to non-WEIRD media systems.
Redefining what “everyday” misinformation even is
A third cluster argues the alarmist literature has been looking in the wrong place entirely. Hourigan2026-oc’s diary study finds that Australians’ lived experience of misinformation is dominated not by exotic AI or conspiracy content but by clickbait-y mainstream and alternative news headlines about business, economics, and celebrity — banal, text-based, and only rarely objectively false when independently verified. Marwick2025-vx challenges the “exposure equals belief” model at the heart of radicalization panic, showing through Far-Right narrative analysis that adopting extremist beliefs is typically a slow socialization process mediated by community and identity, not a single traumatic encounter with disinformation. Prochaska2025-ef similarly argues that election misinformation cannot be assessed post-by-post: much of its persuasive work happens in the surrounding “deep story” context that single-post metrics — the bread and butter of much alarmist quantification — simply cannot see. Frischlich2025-vn ties this together conceptually, arguing that the dominant “infodemic,” “information warfare,” and “pollution” metaphors each illuminate only one level of a genuinely multilevel, complex-adaptive system, and that policy built on any single analogy will misfire.
Complications: where effects and asymmetries do show up
The recalibration is not a blanket null result, and several papers in this collection show where real effects persist, adding nuance rather than contradiction. Adam2026-tz finds genuine, if modest and conditional, contagion and mitigation effects of alternative versus mainstream media on COVID conspiracy beliefs, mediated by selective exposure among the already-populist. Rossini2026-jn finds that belief in electoral misinformation in Brazil’s 2022 election does predict rising political intolerance, and that messaging-app news use has an indirect effect via those beliefs — a reminder that private, encrypted channels (Ventura2025-sw’s WhatsApp deactivation study addresses the same terrain) may behave differently from the public feeds most “no-effects” studies examine. Hameleers2026-mc complicates the generative-AI panic specifically, showing visual disinformation only outperforms text when an issue is salient enough to have plausible existing footage to misappropriate — the effect is real but narrowly contextual. Renault2025-uh documents a robust partisan asymmetry in Community Notes flagging that survives multiple robustness checks, undercutting claims that any observed asymmetry simply reflects fact-checker bias. Van_Erkel2026-mk identifies a “hostile misinformation effect” — partisans perceive their own side as disproportionately targeted by misinformation — that is a perceptual bias with real downstream consequences for trust, even absent objective content asymmetries. Karlsson2026-hd shows that media narratives about interference, rather than interference itself, can independently generate perceived polarization when the “wrong” side is portrayed as beneficiary — suggesting that alarmist coverage can itself become a mechanism of the very polarization it warns against. Green2025-ap and Bakshy2015-rn round out this complicating strand at the level of exposure structure: both show real, asymmetric ideological sorting in what people see and share, even as they push back against strong echo-chamber and filter-bubble claims — a nuance also echoed by Brown2026-br’s YouTube audit, which finds robust content “rabbit holes” but no algorithmic radicalization pathway per se.
Toward a more disciplined research agenda
Several pieces step back to ask how the field should proceed given this recalibration. Spampatti2026-kx imports lessons from climate psychology — behavioral rather than cognitive proxies, cross-national generalizability, systems thinking, wariness of corporate-funded research — as a corrective to a misinformation field still overly reliant on US-centric, belief-level, single-domain measures. Voelkel2026-lc’s megastudy of the ten most-cited climate messages exemplifies the discipline being called for: even well-established persuasive strategies move attitudes only a few percentage points and fail to shift costly behavior, a sober empirical counterpoint to both misinformation and correction-intervention hype. Read together, this set of papers does not deny that misinformation exists or occasionally matters; it insists that its prevalence is modest, its reach concentrated among identifiable subpopulations, its definition broader and more mundane than scholarly attention suggests, and its causal consequences for polarization, radicalization, and democratic health considerably harder to establish than alarmist narratives — in press coverage and in parts of the academic literature alike — have assumed.