Definitional Battlegrounds: What Is “Fake News”?
The foundational move in this literature is a refusal to take “fake news” at face value. Giglietto2019-e9be81c1 argues that creator-centric definitions—those hinging on the intent of whoever fabricated a story—cannot explain how falsehood actually circulates in a hybrid news system; drawing on second-order cybernetics, it relocates the analytic problem from injectors to propagators, whose successive judgments compose the false-information cycle. Freelon2020-yp performs a parallel move at the level of vocabulary, proposing that “disinformation” (deception + harm + intent) supplant “fake news” as an analytical category precisely because Trump-era politics stripped the latter of definitional value—a claim borne out empirically by Yoo2026-ev, which shows both left- and right-wing outlets using “fake news” almost entirely as a delegitimizing slur rather than a description of fabricated content, and by Mosca2026-yh, whose “Strategic Delegitimization and Selective Amplification” model treats fake-news accusations themselves as a rhetorical weapon in negative campaigning. Bennett2018-ys pushes the diagnosis further outward, arguing that disinformation is symptomatic of a systemic “disinformation order” rooted in declining institutional trust rather than a discrete media-technological pathology—an argument that dovetails with Chadwick2013-ns and Chadwick2015-ii, whose “hybrid media system” concept (interacting old/new media logics, relational power, political information cycles) supplies much of this corpus’s underlying vocabulary for how falsehood moves across, rather than within, single platforms or genres.
A second wave of conceptual work interrogates the adequacy of the “information disorder” frame itself. Frischlich2025-vn shows that the dominant analogies—infodemic/virus, information warfare, information pollution—each illuminate only one level of a multilevel complex system and can license counterproductive interventions. Thiele2025-ol and Mannocci2026-kc perform similar conceptual audits of “coordinated inauthentic behavior,” arguing that platform-derived definitions conflate inauthenticity with manipulative intent and are analytically too thin to ground either detection or normative judgment; both propose more theory-driven typologies (a rational-choice principal model in the former, an actors/actions/intent framework in the latter) as correctives. Dierickx2026-tw extends the definitional project into the generative-AI era, arguing that conventional fact categories (evidence-based, interpretative, rule-based) cannot capture AI outputs and proposing “emergent facts” as a new epistemic category—an update of the same foundational impulse that animated Giglietto’s critique a half-decade earlier: existing typologies are structurally unfit for the systems they claim to describe.
The Narrative Turn
A cluster of papers relocates the unit of analysis from discrete false claims to stories. Marwick2026-qd argues disinformation is best conceptualized as cross-platform, identity-affirming, cathartic narrative, with catchphrases and visual motifs (rather than domain-level classification) as the operative unit of spread. Sadler2025-vu pushes this further via hermeneutic narratology, showing that content can be referentially and ethically problematic without being straightforwardly false, thereby destabilizing the true/false binary that undergirds most fact-checking paradigms. Starbird2025-jj and Prochaska2025-ef operationalize this narrative turn empirically on election-integrity discourse, showing that misleading rumors emerge from the interaction of (often accurate) evidence with partisan frames, and that “deep stories” are performed collaboratively across influencers, elites, and audiences over years, not single posts. Goel2025-iq demonstrates the mechanism at scale: mainstream, factually true articles are disproportionately co-shared with fake news to support misleading narratives, meaning source-level blocklists systematically undercount the phenomenon. Suau_Martinez2026-lv confirms the belief-formation payoff of this reframing, showing cumulative narrative exposure—not isolated claims—drives belief. Together these works constitute a coherent rebuttal to information-centric, claim-by-claim models of disinformation, of a piece with Marwick2025-vx’s account of “processual redpilling,” where extremist belief is less a single false claim internalized than a socialization process in which disinformation functions as accumulated “evidence.”
Was the Alarm Overblown? The Prevalence Debate
Running parallel to (and often in tension with) the conceptual literature is an empirical corrective tradition insisting that “fake news” alarm has outrun the evidence. Allcott2017-yz, Grinberg2019-ua, Guess2019-ym, Guess2020-rr, Guess2021-ym, Allen2020-nj, and Bergeron-Boutin2026-yh converge on a common finding: exposure to and sharing of low-credibility content is small relative to total media diets, heavily concentrated among a narrow, older, more conservative subpopulation, and generally not the crowding-out force popular narrative assumes. Allcott2019-gn tracks the decline of this problem on Facebook after 2016 even as it persisted on Twitter, while Budak2024-ef synthesizes this whole tradition into an explicit argument that public discourse has diverged from scientific evidence on prevalence, algorithmic causation, and societal harm. Vosoughi2018-at complicates the picture by showing falsehood genuinely out-diffuses truth on Twitter—an effect attributable to human sharing behavior (novelty-seeking) rather than bots—while Yang2023-cg argues that link-based measures used throughout this “it’s-overblown” literature systematically miss a far larger reservoir of image-based misinformation, heavily skewed to the political right. Nenno2025-xa and Luhring2025-od extend the methodological self-scrutiny cross-nationally, the former showing user-flagged misinformation is only modestly distinguishable by news values across WEIRD/non-WEIRD contexts, the latter auditing NewsGuard itself and showing that binary trustworthy/untrustworthy classifications (rather than continuous scores) can substantially distort downstream findings—a caution that retroactively qualifies much of the domain-list-based prevalence literature.
Elite Strategy, Polarization, and the Politics of Falsehood
A distinct strand insists disinformation is best explained as elite political strategy rather than mass cognitive failure or platform architecture. Benkler2018-lw’s “network propaganda” thesis—an asymmetric, radicalized right-wing media ecosystem insulated from truth-correcting norms—anchors this line, echoed and extended by Tornberg2025-ir, which finds cross-nationally that neither populism nor right-wing ideology alone predicts elite misinformation-sharing, but their combination (radical-right populism) does. Gaber2022-bk theorizes “strategic lying” as a deliberate agenda-setting tactic evolved from ordinary political spin, using Brexit and the 2019 UK election to show how rebuttals and fact-checks can inadvertently amplify the very lies they target. Tai2026-qk and Rodarte2026-dk look at elected officials directly: the former testing institutional/ideological checks on politicians’ misinformation-sharing, the latter showing Brazilian parliamentarians pursuing competing modes of “epistemic authority” (derivative reporting, direct authority claims, proximity-based brokerage) during a health crisis—effectively becoming media producers competing with journalism for gatekeeping power. Bennett2025-xs synthesizes institutionalist and technocentric accounts of democratic backsliding via “digital surrogate organizations,” arguing that platform-enabled organization, not just persuasion, is the missing mechanism connecting online disinformation to elite radicalization. At the level of citizen psychology, Osmundsen2021-et and Pennycook2021-jq debate whether partisan hostility or simple inattentiveness better explains fake-news sharing, while Iyengar2019-jj supplies the affective-polarization backdrop against which Van_Erkel2026-mk’s “hostile misinformation effect” and Lewandowsky2012-vn’s classic account of the “continued influence effect” operate—foundational psychological mechanisms that this topic’s later empirical papers (on Brazil, on Europe) repeatedly invoke to explain why corrections fail and disinformation belief tracks partisan identity (Rossini2026-jn, Rossini2026-mj).
Coordinated Inauthentic Behavior and the Attribution Problem
Where the narrative-turn papers ask “what is the story,” a companion cluster asks “who is behind it and how do we know.” Starbird2019-qv reframes information operations as collaborative work, arguing CSCW’s sociotechnical lens exposes how orchestrated campaigns are inseparable from organic crowd participation—undermining simple bot/troll dichotomies. Keller2019-nk’s ground-truthed study of the South Korean NIS astroturfing campaign operationalizes this insight, showing coordination signatures (not bot-like individual traits) are the reliable tell, while finding that even well-resourced human-staffed campaigns had limited measurable discursive impact. Poliakoff2026-fa’s labor-market analysis of the Internet Research Agency similarly de-exoticizes “troll factories,” showing the IRA operated as an ordinary, if covert, PR/media company integrated into Russia’s broader media labor market. Thiele2025-ol and Mannocci2026-kc then formalize the attribution and detection problem theoretically, while Rogers2026-cy operationalizes a related concept—“post-truth spaces”—as an applied tool for fact-checkers to locate influential unverifiable clusters, tested on Moldovan Facebook discourse about the Russo-Ukrainian war. FitzGerald2025-nv and Bosch2024-hj extend this operational lens to campaign persistence over time and platform-specific (sonic) propaganda techniques on TikTok, while Makeev2026-ma theorizes a “dual-objective” propaganda model distinguishing domestic from foreign-facing state messaging in wartime.
Platform Governance, Data Access, and the Politics of Accountability
A large cluster treats platforms themselves—their moderation choices, disclosure practices, and control over research access—as the central object of “information disorder” scholarship. Donovan2025-ws historicizes “misinformation-at-scale” as a structural feature of engagement-driven business models, tracing platforms’ shift from non-intervention to behavioral moderation (coordinated inauthentic behavior) and their retreat after political backlash. Gillespie2022-jx identifies “reduction”—algorithmic demotion short of removal—as an under-theorized but now-dominant moderation modality that evades transparency regimes built around removal. Hurcombe2025-cs performs discourse analysis on Meta’s own Newsroom communications, showing the company frames problematic content as originating from outside actors while excluding its own platform affordances (e.g., Groups) from scrutiny. Cazzamatta2026-lo and Farkas2026-lr turn to fact-checkers’ own position within this system, the former documenting that content removal after verification is in fact rare (undermining Zuckerberg’s censorship framing), the latter showing fact-checkers rhetorically defend platform-funding partnerships via apologia strategies of differentiation and transcendence. Holt2026-zq complicates the post-fact-checking “wisdom of crowds” turn by showing user “False News” reports on Facebook mostly flag polarized-but-true content people disagree with, not falsity per se—a finding with direct bearing on Meta’s 2025 shift to Community Notes. Data-access infrastructure itself becomes an object of study in Bruns2019-nr’s “APIcalypse,” Munger2025-cz’s meta-scientific critique of the Meta2020 partnership’s limited external and “temporal” validity, and the trio de-Vreese2026-zx, Giglietto2026-855a54cb, Pierri2026-ib, which track the fragile, contested emergence of the EU Digital Services Act’s Article 40 as a legal (rather than philanthropic) right to platform data—alongside Crosset2026-mq’s comparative analysis of EU/US content-moderation law as converging on “security bureaucratisation” and Bechmann2026-dr’s call to study platform collective behavior as democratic infrastructure rather than isolated causal effects. Lewandowsky2026-ob situates all this within a broader argument that platforms bear direct responsibility for democratic backsliding.
Comparative Scope and Infrastructural Correctives
Several papers push back against the field’s WEIRD- and US-centrism. Humprecht2020-gd proposes a cross-national resilience framework, clustering eighteen democracies by structural vulnerability and isolating the US as a uniquely fragile outlier; Humprecht2025-ml and Falkenberg2026-ka make the parallel meta-critique explicit for misinformation and polarization research respectively, calling for de-Westernized, integrated agendas. Zhu2026-tn and Heft2021-ky supply comparative network evidence on alternative-media audience overlap and transnational right-wing hyperlink ecologies, showing “alternativeness” is relationally and institutionally contingent rather than fixed. Gattermann2025-yx connects far-right electoral performance to voter disinformation concern across the 2024 European Parliament elections. Methodological infrastructure receives its own foundational treatment: Scalco2026-bd operationalizes “information voids” as measurable supply/demand anomalies, Van-Eck2026-xg’s (van-Eck) review shows depolarization recommendations are typically disconnected from the polarization evidence researchers themselves generate, and Mahl2026-hc’s Delphi study of 47 experts maps the field’s fragmentation and anticipates future problem areas (notably, generative AI was not foreseen as a major factor as late as 2022). Spampatti2026-kx imports lessons from the more mature field of climate psychology—on intervention design, behavioral measurement, and researcher self-organization against vested-interest attacks—as a template for maturing misinformation research itself.
The Generative AI Frontier
The corpus’s most recent stratum extends these foundational categories into the era of synthetic media. Emilio2026-ik formalizes “synthetic reality” as a four-layer stack (content, identity, interaction, institutions) and names the “Generative AI Paradox”—societies may rationally discount all digital evidence as synthetic media proliferates, raising the cost of truth itself. Triedman2025-uy empirically substantiates this concern via a comparative audit of Grokipedia against Wikipedia, finding heavier reliance on blacklisted sources and ideological skew on controversial topics. DeVerna2025-dl shows that even reasoning- and search-enabled LLMs fact-check poorly absent curated retrieval, while exhibiting citation bias toward left-leaning sources. Kasianenko2026-tn extends algorithmic auditing to AI Overviews’ handling of conspiratorial search queries, finding uneven guardrails that can inadvertently legitimize newer conspiracy narratives. Choi2026-bz and Xue2025-bp probe downstream psychological effects—modality-congruent carryover in deepfake authentication difficulty, and the strategic use of emotionality in fact-checking messaging—while Ramos2026-qo proposes “argument-checking” and narrative monitoring as necessary complements to claim-level fact-checking for the crisis-driven, rapidly mutating disinformation narratives (e.g., the 2024 Valencia floods) that generative tools now accelerate. Read together with Gonzalez-Bailon2024-rq and Del-Vicario2016-uj on diffusion dynamics, this frontier suggests the field’s foundational task—defining what counts as false, whose intent matters, and at what level of analysis disorder should be measured—remains as unsettled for AI-generated content as it was for the original 2016-era “fake news” debates that opened this note’s arc.