Information Disorder and Fact-Checking
The Concept Under Strain
The papers gathered here register a field in the midst of conceptual self-examination. Several contributions argue that core terms — misinformation, fake news, narrative, fact — have become analytically blunt. Frischlich2025-vn dissects the dominant analogies (infodemic, information warfare, information pollution), showing that each carries productive “conceptual mileage” alongside obscuring “baggage.” Sadler2025-vu argues that disinformation studies has used “narrative” intuitively, defaulting to true/false binaries that miss how stories disclose rather than represent. Dierickx2026-tw pushes further still, proposing “emergent facts” as a new epistemic category for GenAI outputs that established positivist, constructivist, and institutional frameworks cannot accommodate. Yoo2026-ev documents the parallel hollowing-out from below, showing how “fake news” has been weaponised across the ideological spectrum into a mere delegitimation label. Together, these papers signal that the field’s vocabulary is being renegotiated under empirical and political pressure.
This reconceptualisation is reinforced by audits of measurement infrastructure. Luhring2025-od shows that source-based ratings via NewsGuard are stable in continuous form but produce wildly different results when binarised. Goel2025-iq demonstrates that domain-list approaches miss a perhaps larger phenomenon: factually accurate mainstream content repurposed into misleading narratives. Waight2025-al proposes LLM-based narrative-similarity measurement that catches Russian–mainstream narrative overlap that ngram methods render invisible. The pattern: how we measure determines what counts as misinformation, and most established estimators understate the problem in some directions while overstating it in others.
What Is Actually Going On Out There?
A second cluster offers sober prevalence estimates that cut against alarmist public discourse. Budak2024-ef argues directly that public narratives overstate exposure, algorithmic causation, and societal effects. Gonzalez-Bailon2024-rq and Appel2026-vz — sorry, Appel2026-qr — find that on Facebook, deceptive content reaches many but represents only ~0.3% of political content views, with effects largely vanishing after controlling for pre-exposure characteristics. Hourigan2026-oc finds via digital diaries that users perceive misinformation largely in mainstream outlets via clickbait, and that only 9% of flagged claims were objectively false. Mosleh2024-op shows that single-platform studies (mostly Twitter) badly misrepresent the broader ecosystem.
Yet other papers complicate this deflationary turn. Vincent_undated-re reports TikTok mis/disinformation prevalence rising to ~25%. Bollenbacher2026-vz uses a SIRVA epidemic model to link antivaccine tweets causally to ~14,000 vaccine refusals and excess deaths. Kim2026-wg documents geo-racially targeted voter suppression ads depressing turnout by up to 17% in targeted segments — effects invisible to average-treatment-effect designs. Rossini2026-jn traces how electoral misinformation eroded political tolerance in Brazil over a three-wave panel. The reconciliation is methodological: average exposure is low, but targeted and narrative-level effects on vulnerable populations or institutions can be severe.
The Ecology of Production
Several papers shift focus from individual content to the institutional machinery producing it. Poliakoff2026-fa uses 350 CVs to reconstruct the IRA as a labour-market actor with a five-level hierarchy, integrated into Russia’s media economy — not a sui generis covert operation. Thiele2025-ol formalises this turn with a rational-choice typology of coordinated manipulation campaigns. Kansaon2025-id documents coordinated WhatsApp activity translating into the January 8 Brasília attacks. Copland2025-em traces how Sky News Australia became a node in transnational right-wing network propaganda. Donovan2025-ws argues that “misinformation-at-scale” is a structural feature of engagement-driven business models, not a bug — a thesis given uncomfortable empirical weight by Waight2026-ts, which shows state-coordinated Chinese media seeping into LLM training data and shifting commercial model outputs.
Political elites themselves are increasingly the producers. Rodarte2026-dk shows Brazilian parliamentarians contesting epistemic authority through three distinct modes during the Manaus crisis. Tai2026-qk examines institutional and ideological checks on elected-official misinformation sharing. Renault2025-uh finds Republicans flagged 2.3 times more often than Democrats even on Community Notes — undermining the partisan-bias justification for dismantling professional fact-checking. Bennett2025-xs synthesises these threads into a “digital surrogate organizations” framework, arguing that platform-enabled organisation — not just persuasion — is the missing link between media effects and democratic backsliding.
Audiences, Belief Formation, and Mechanism
A robust mechanistic literature continues to map how exposure becomes belief. Adam2026-tz combines panel surveys with web tracking to demonstrate both contagion (alternative media) and mitigation (mainstream debunking) effects on COVID conspiracy beliefs, mediated by populism and political mistrust. Rohrbach2026-rc examines how journalistic strategies themselves contribute to conspiracy formation. Di-Domenico2026-zq reframes online toxicity as source-amplified, with influencer credibility legitimising misinformation through parasocial enmeshment. Marwick2025-ov reads ConspiracyTok as “populist knowledge production” by young, non-White, female creators — challenging the stereotype of conspiracy belief as a White-male pathology.
The narrative turn pushes belief formation past discrete claims. Starbird2025-jj and Prochaska2025-ef develop evidence-frame and deep-story frameworks showing that quote-tweet structures and elite cues let audiences supply misleading frames around factually accurate evidence. Gardam2025-er pursues a parallel visual-first methodology for Instagram climate denial. Hameleers2026-mc complicates deepfake panic by showing that AI-generated images do not outperform low-tech decontextualised video — and only for high-salience issues with ample authentic footage.
Counter-Strategies and Their Limits
The counter-measures literature has matured from optimism to realism. van-der-Linden2026-jt demonstrates that prebunking ads can be scaled cheaply on Instagram with persistent 5-month effects. Lieu2025-nl tests CARDS and FLICC taxonomies on climate misinformation, finding content category matters more than logical fallacy type. Choi2026-bz surfaces modality-congruent carryover effects in deepfake exposure. Xue2025-bp examines how fact-checkers leverage emotionality — uncomfortable for the genre’s objectivity norms.
But fact-checking is now in institutional crisis. Farkas2026-lr (and its near-duplicate Shi2026-ko) document European fact-checkers’ apologia rhetoric defending platform partnerships. Cazzamatta2026-lo shows that Meta’s January 2025 termination of its Third-Party Fact-Checking Program rests on empirically unfounded censorship claims — content removal occurs in only ~30% of cases. Triedman2025-uy audits Grokipedia as a politically inflected AI alternative to Wikipedia, citing Stormfront and InfoWars at rates that suggest a deliberate ideological rewrite. DeVerna2025-dl shows that LLMs can fact-check well only with curated RAG context — automation will not replace professional infrastructure. Mahl2026-hc’s Delphi study summarises the expert consensus: governance and journalism interventions outrank individual-level fixes, and there is no silver bullet.
Comparative and Global Reorientation
Finally, the field is decisively de-Westernising. Humprecht2025-ml frames a special issue around comparative work beyond WEIRD contexts. Nenno2025-xa examines news values across 24 countries, finding the US and Brazil as polarised outliers. Gattermann2025-yx links far-right electoral performance to voter disinformation concerns across the EU. Ventura2025-sw uses a WhatsApp deactivation experiment in Brazil to push beyond feed-based platforms. Bosch2024-hj foregrounds sound on TikTok as platform-specific propaganda. Spampatti2026-kx urges misinformation researchers to learn from climate psychology’s experience with organised denial, including resilience to coordinated attacks on researchers themselves. Emilio2026-ik articulates the broader stakes via the “Generative AI Paradox”: as synthetic media proliferates, rational actors may discount all digital evidence, raising the cost of truth itself. FitzGerald2025-nv and Graham2025-gp examine how manipulation campaigns persist by appropriating events and exploiting platform infrastructure.
The arc of this body of work runs from definitional crisis through institutional analysis to chastened policy realism. Where the field once treated misinformation as discrete false claims circulating among gullible individuals, it increasingly treats it as a narrative, infrastructural, and institutional phenomenon — produced by political and economic systems, contested through epistemic boundary-work, and resistant to single-lever solutions.