The Shape of the Field: From Content to Behavior to Impact
This is the researcher’s deepest and longest-running vein of work, and the papers filed here trace a coherent intellectual arc: from establishing behavior rather than content as the proper unit of analysis for platform manipulation, through building and refining detection methodologies, to interrogating whether coordination actually matters for opinion change — and finally confronting a new generation of AI-driven threats.
Foundations: coordination as an ecological signal
The line of argument begins with Giglietto2020-9d8acdd7, which reframes disinformation research away from adjudicating truth/falsity or hunting “bad actors” toward detecting coordinated collective action — operationalizing what Facebook had called “coordinated inauthentic behavior” using near-simultaneous, repeated link-sharing across pages, groups, and profiles during Italian elections. This CooRnet/CLSB methodology becomes the backbone of an entire research program, extended into pandemic-era monitoring in Giglietto2022-0e951ac5 (the MINE-FACTS workflow feeding Italian fact-checkers), into dynamic list-updating in Giglietto2023-fa71a001 (2022 Italian snap election, surfacing hyperpartisan, click-economy, and religious-proselytism networks), into influencer ecosystem analysis in Marino2023-9137f448 (Alternative Influence Networks and the “Intellectual Dark Web” remediated through legacy media), and most recently into visual/generative-AI persuasion in Giglietto2026-9b6a992d (gambling promotion networks) and the broader synthesis in Giada2026-fc9a3833, which proposes “Deceptive Information Operations” as a category that subsumes CIB, deception theory, and strategic information operations under a single structural logic. Parallel conceptual work from Starbird’s tradition — Starbird2019-qv — insists disinformation is collaborative work, blurring orchestrated and organic participation, a theme echoed decades later in Starbird2025-jj’s evidence-frame analysis of election-integrity rumoring, where misinformation is shown to live in the interaction between accurate evidence and partisan framing rather than in false facts per se.
Methodological refinement: what counts as coordination?
A second cluster interrogates and extends the CLSB/coordination-detection toolkit itself. Nizzoli2020-cf argues coordination is a continuum, not a binary, and demonstrates it is orthogonal to automation — a claim central to the field’s self-understanding. Keller2019-nk uses rare ground-truth data from South Korea’s NIS scandal to show coordination traces, not bot-like features, are the most reliable signal, while also introducing a principal-agent lens later formalized by Thiele2025-ol, whose typology of eight campaign types tackles the “attribution problem” — who commissions manipulation — as analytically distinct from detecting inauthenticity. Technical refinements proliferate: Mannocci2025-ig and Iannucci2025-eg both push toward multimodal and temporal-aware detection, showing single-modality or flattened-network approaches miss or dilute real coordination; Yang2025-iv proposes evasion-resistant statistical signals of link-sharing speed/frequency; Minici2024-tf introduces a graph foundation model (IOHunter) for cross-campaign generalization; Rodriguez_Farres2025-sg and Luceri2025-tr extend detection to new platform contexts (Bluesky, TikTok) where text-centric assumptions break down. Mannocci2026-kc caps this thread with a PRISMA survey proposing a unifying definition built on actors, synergic actions, and intent, explicitly decoupling coordination from harm. Skepticism about interstate and platform-based coordination claims appears in Pante2025-pq, which re-tests prior inter-state collaboration claims against control datasets.
Does coordination matter? The causal reckoning
A striking counter-current questions whether detected coordination and exposure actually shift attitudes or behavior. Eady2023-xg finds IRA exposure in 2016 was concentrated among already-partisan Republicans and dwarfed by domestic media exposure, with no measurable attitudinal effect — a finding structurally paralleled by Appel2026-qr’s FIES analysis of 49 deceptive 2020 networks, where correlations with discernment and polarization evaporate after controlling for pre-exposure characteristics. Di-Marco2025-aa and its companion Di_Marco2026-xu formalize this skepticism graph-theoretically, showing real coordinated accounts in UK election cascades achieve influence far below optimal or even random-matched benchmarks. Against this, Kim2026-wg recovers a strong causal signal by attending to heterogeneous, geo-racially targeted effects invisible to average-treatment-effect designs, and Bollenbacher2026-vz links antivaccine Twitter exposure causally to compartmental epidemic outcomes. Karlsson2026-hd adds a further twist: even when interference doesn’t change votes, media narratives about interference can independently polarize audiences along partisan lines.
Domain and country case studies
The topic’s comparative breadth is anchored by case studies spanning the named countries and beyond. Italy dominates via the CooRnet lineage above; health misinformation is examined comparatively in Song2025-yh (UK/US anti- and pro-vaccine CLSB) and Efstratiou2026-ij (COVID science communication pathways); Brazil appears via Kansaon2025-id’s WhatsApp coordination analysis linking online mobilization to the January 8 attacks; Nigeria and India are less directly represented here but sit within the broader Global South brokerage frame of Gaw2025-ru (Philippines) and the electoral-authoritarian frame of Jovanovic-Harrington2026-ze (Serbia); Ukraine/geopolitical cases include Waight2025-al’s narrative-similarity tracing of Russian “biolabs” claims, Graham2025-gp’s #IStandWithPutin case study, Bosch2024-hj on TikTok sound-based Ukraine disinformation, and Rogers2026-cy’s “post-truth spaces” mapping of Moldovan Facebook discourse. Russian and adjacent state operations are historicized organizationally in Poliakoff2026-fa’s OSINT reconstruction of the IRA’s labor structure, and extended to Telegram in Kuznetsova2025-nu (Russia/Belarus) and to Korea in Kim2026-br’s two-decade troll-detection study. China’s protest-repression dynamics appear in Kulichkina2026-zk.
Adjacent phenomena and platform politics
Several papers push the topic’s boundaries productively. Simeone2025-vo shows deplatforming produces network-wide “ripple effects” independent of CIB per se; Graham2026-fb and Bastos2025-ol interrogate the human/bot and visual-identity boundaries that underpin detection assumptions; Zhao2025-ny and Arceneaux2026-xk examine bots as agenda-builders and hashtag-hijackers in more conventional political communication terms; Holland_Levin2026-qx, Knupfer2025-vt, Copland2025-em, and Bouchafra2026-ts trace cross-platform politicization and propaganda dynamics (culture-war discourse, connective factions, Sky News Australia, Sweden Democrats) that intersect with but aren’t reducible to coordination detection. Gerard2025-br offers a structural innovation — narrative-cluster embeddings identifying “bridge users” who seed cross-platform migration between Truth Social and X. Jurg2025-ur and Oprea2025-lf turn to platform accountability, auditing YouTube’s moderation and documenting unaddressed hyperactive-user amplification on Facebook during EU elections. Di-Domenico2026-zq extends coordination-adjacent thinking into marketing, showing influencer-sourced misinformation generates distinct toxic response patterns.
The AI horizon
Finally, a forward-looking cluster confronts generative AI as both a subject and a tool of coordinated manipulation. Orlando2025-ul simulates emergent LLM-agent coordination, showing mere mutual awareness among synthetic agents approximates explicit collusion; Schroeder2026-im extrapolates this into a governance-oriented warning about “malicious AI swarms”; and FitzGerald2025-nv notes the adaptive persistence of manipulation campaigns across emerging events — together suggesting that the CooRnet-era behavioral signatures this research program pioneered will need continual reinvention as the actors themselves become synthetic, adaptive, and increasingly difficult to disentangle from authentic participation.