From false facts to narrative: a conceptual turn
The thread that ties this collection together is a shared discontent with the “true/false” binary that dominated first-generation disinformation research. Giglietto2019-e9be81c1 opens this argument early, showing via second-order cybernetics that intent-centric definitions of “fake news” cannot capture how information actually cycles through a hybrid media system — attention shifts instead to propagators and the meso-level judgments that turn a claim into misinformation, disinformation, or something composite. Marwick2026-qd, Sadler2025-vu, and Suau_Martinez2026-lv push this insight further by arguing that disinformation is best studied as narrative: cross-platform, identity-affirming, cathartic stories (Marwick), hermeneutically “disclosive” rather than simply false accounts (Sadler), or cumulative exposure effects that build belief over repeated tellings (Suau Martinez). Prochaska2025-ef and Starbird2025-jj operationalize this narrative turn empirically, showing that U.S. election-fraud rumors are less about discrete false facts than about “evidence-frame” dynamics and “deep story” performances co-produced by influencers, elites, and audiences across quote-tweets and comment threads. Goel2025-iq extends the argument in an unexpected direction: even scrupulously accurate mainstream articles are disproportionately co-shared to prop up misleading narratives (on vaccines, on voter fraud), meaning that source-level “fake news” metrics dramatically undercount the phenomenon. This cluster converges on a methodological consequence taken up directly by Waight2025-al and Ramos2026-qo: if narrative rather than falsity is the operative unit, then detection and correction tools need to measure narrative similarity and argumentative structure (premises, fallacies) rather than only fact-check individual claims — Waight’s LLM-based narrative-similarity pipeline and Maldita.es’s “argument-checking” workflow (built around the Valencia DANA flood conspiracy) are direct methodological responses to this reframing.
Elite strategy and the politics of misinformation
A second cluster treats misinformation and “fake news” rhetoric as elite political strategy rather than epistemic accident. Rodarte2026-dk shows Brazilian parliamentarians contesting journalistic gatekeeping during the Manaus COVID collapse through distinct epistemic-authority repertoires; Rossini2026-mj extends the Brazilian case to Bolsonaro’s systematic assault on electoral legitimacy, finding mainstream news use bolsters trust even among his own voters while belief in fraud narratives corrodes it. Tornberg2025-ir generalizes this across 26 countries: neither populism nor right-wing ideology alone predicts elite misinformation sharing, but their conjunction — radical-right populism — does, tying misinformation to a specific ideological-institutional posture rather than a generic information disorder. Yoo2026-ev and Mosca2026-yh examine the discursive weaponization of the label itself, showing how “fake news” accusations function rhetorically to redraw Hallin’s spheres of consensus and deviance (Newsmax vs. Occupy Democrats) or to structure negative campaigning and selective media amplification in Italy. Tai2026-qk rounds out this cluster by asking which institutional and ideological positions act as checks on elected officials’ own misinformation-sharing — a question that dovetails with Tornberg’s party-level findings.
The Ukraine war as narrative battleground
Several papers converge on the Russo-Ukrainian war as an especially rich site for observing narrative propagation and propaganda design. Sadler2025-vu and its narratological toolkit dissect a Venezuelan influencer’s NATO-blame narrative to show how content can be referentially and ethically suspect without being straightforwardly false. Waight2025-al traces the “Ukrainian biolabs” claim’s diffusion from Russian state media into low-quality and even mainstream U.S. outlets, demonstrating empirically that lexical-overlap methods badly undercount narrative migration. Graham2025-gp and Bosch2024-hj examine platform-specific propaganda mechanics — hashtag/verification infrastructure exploitation in IStandWithPutin, and sonic affordances in TikTok’s Ukraine ecosystem. Rogers2026-cy, conducted under the EU vera.ai project, maps Moldovan Facebook’s Russia-Ukraine discourse through “post-truth spaces,” finding that a “borderline” cluster bridging fringe and mainstream content — rather than fringe sources themselves — drives the war conversation. Giada2026-fc9a3833 includes a pro-Putin coordinated network as one of its three “Deceptive Information Operations” case studies, and Makeev2026-ma theorizes the dual domestic/foreign propaganda logic underlying Russian state messaging about the war more broadly.
Coordinated inauthentic behavior, attribution, and platform governance
A methodologically dense cluster, closely aligned with the PROMPT and vera.ai monitoring agenda, investigates the organizational and infrastructural mechanics of coordinated manipulation. Poliakoff2026-fa offers a rare organizational-labor account of the IRA “troll factory,” showing it functioned as a conventional media/PR employer embedded in Russia’s labor market rather than a shadowy intelligence unit. Kuznetsova2025-nu and Zhao2025-ny examine coordination mechanics on Telegram (Russia/Belarus regime amplification) and hashtag hijacking dynamics involving bots and narrative substitution, while Kim2026-br applies explainable ML to two decades of Korean news comments, finding that suspected trolls favor morally condemning rhetoric over direct praise — a subtler and more algorithmically rewarded strategy than naive boosting models predict. Thiele2025-ol steps back to theorize the field’s core “attribution problem,” proposing a rational-choice typology linking campaign scale, elaborateness, and disguise to principals’ resources and stakes — explicitly critiquing Meta’s “coordinated inauthentic behavior” category as too narrow. That critique is empirically vindicated by Giada2026-fc9a3833 and Giglietto2026-9b6a992d, both emerging directly from PROMPT/vera.ai monitoring infrastructure: the former’s “Zombie Internet” concept and three Facebook case studies (pro-Putin propaganda, gambling promotion, unmoderated adult-content groups) show that CIB/CSH enforcement frameworks miss manipulation whose harm lies in the convergence of deceptive content, affordance exploitation, and coordination rather than in any single dimension; the latter documents an exponential, ChatGPT-triggered explosion in AI-generated gambling imagery exploiting Meta’s paid/organic regulatory gap. FitzGerald2025-nv and Gaw2025-ru extend the infrastructural argument temporally and institutionally — manipulation campaigns persist by appropriating new events, and Philippine influence operations function as durable political-economic brokerage rather than ad hoc trolling.
Platform ecosystems, mainstreaming, and audience culture
A related set of papers studies how platform affordances and audience cultures channel extremist or securitizing content into the mainstream. Rothut2026-or develops “protest-facilitated mainstreaming” from the German Querdenken movement’s Telegram entanglement with far-right networks; Copland2025-em traces Sky News Australia’s transformation into a node of transnational right-wing “network propaganda” via a dual Facebook/YouTube strategy; Bouchafra2026-ts and Hollingshead2026-vx compare platform-tailored visual securitization (Sweden Democrats’ EU campaign) against surprisingly pro-immigration TikTok vernacular in Canada, both stressing that platform affordances are used symmetrically across ideological positions. Marwick2025-ov complicates stereotypes further, showing ConspiracyTok’s “generous epistemology” is driven by young, non-white, and female creators performing “populist knowledge production.” Di-Domenico2026-zq shows influencer-sourced misinformation generates qualitatively distinct, more homogeneous toxicity than identical claims from regular users, implicating credibility and parasocial bonds as amplification mechanisms in their own right.
Generative AI: persuasion, propaganda, and epistemic security
The newest and fastest-growing cluster addresses generative AI’s role in both interpersonal persuasion and industrial-scale information operations. On persuasion, Hackenburg2025-dj, Hackenburg2026-ud, and Lin2025-xp form a tight empirical program: conversational AI outpersuades even elite human debaters and professional canvassers, with information density (fact-checkable claims per exchange) — not personalization or model scale — identified as the operative mechanism, alongside a troubling persuasion-accuracy trade-off and a systematic rightward-skewing inaccuracy asymmetry. Waight2026-ts shows this fact-density story has a geopolitical shadow: state-coordinated media memorized in training corpora make LLMs more pro-regime in languages associated with lower media freedom, a subtle “laundering” pathway distinct from direct model manipulation. At the operational end, Schroeder2026-im and Emilio2026-ik theorize the frontier risk — malicious multi-agent “AI swarms” manufacturing consensus, and a four-layer “synthetic reality” stack producing an “epistemic tax” that may lead societies to rationally discount all digital evidence. Schiffrin_undated-gi grounds this in concrete harm and policy response, mapping the deepfake financial-fraud ecosystem and arguing for gatekeeper-liability regulation over victim-blaming frameworks. Wack2026-bt provides crucial audience-reception evidence from Kenya, showing deepfake impact is not fixed at production but socially negotiated through surrounding comments and partisan motivated reasoning — directly relevant to Global South contexts underserved by moderation and labeling infrastructure. Beacken2026-zb insists this uptake is not uniform, comparing GenAI propaganda across six democratically weakened states to argue against technologically deterministic narratives. Hameleers2026-mc offers an important corrective note to the generative-AI alarm: decontextualized real video, not AI-generated stills, proved most credible and engaging disinformation for the salient Ukraine-war issue, suggesting low-tech recontextualization remains at least as potent as synthetic media. Giglietto2026-9b6a992d and Giada2026-fc9a3833 tie this cluster back to the coordinated-behavior literature empirically, documenting how generative AI intensifies (rather than replaces) established persuasive visual drivers in industrial-scale gambling promotion.
Synthesizing the field
Sitting above all these strands, Mahl2026-hc’s Delphi study with 47 international experts offers a field-level anticipatory assessment that anticipates much of this note’s own architecture: it calls for integrative, multi-level, multi-stakeholder interventions across platform governance and journalism ecosystems rather than siloed, individual-focused fixes — and notes, tellingly, that generative AI’s disruptive potential was barely anticipated as recently as 2022. Read together, these forty-five papers trace an arc from information-centric, fact/falsity paradigms toward narrative-, infrastructure-, and now generation-centric understandings of disinformation — a trajectory the PROMPT project’s own monitoring tools (Giglietto2026-9b6a992d, Giada2026-fc9a3833, Rogers2026-cy, Waight2025-al) are actively built to keep pace with.