The line of argument: from behavioral signature to sociotechnical infrastructure
The thread running through this collection begins with a methodological wager: that inauthentic influence is best detected not through content veracity or actor identity, but through the behavioral signature of coordination itself. Giglietto2020-9d8acdd7 establishes this founding move, arguing that disinformation research should shift from adjudicating truth or unmasking bad actors toward detecting coordinated collective action—operationalizing this as “coordinated link sharing behavior” (CLSB) on Facebook and showing its correlation with problematic domains during Italian elections. This paper is the conceptual seed of CooRnet, and the corpus around it traces how that seed grows into an entire research program. Giglietto2022-0e951ac5 extends the method into COVID-19 fact-checking, showing CLSB can outperform manual fact-checking in surfacing problematic content and cataloguing evasion tactics (link laundering, first-comment link-hiding, religious-group cover). Giglietto2023-fa71a001 then solves a structural weakness of static lists—their rapid obsolescence—by proposing a continuously updating workflow, applied to the 2022 Italian election and yielding case studies spanning political, click-economy, and religious-proselytism operations. This lineage culminates in explicitly VERA-AI-affiliated work: Giglietto2026-9b6a992d applies coordinated-network detection to visual persuasion in AI-generated gambling promotion, documenting an exponential post-ChatGPT surge in synthetic content; Giada2026-fc9a3833 synthesizes three such VERA-AI-detected operations (pro-Putin propaganda, gambling, and adult-content/fraud networks) into the broader concept of “Deceptive Information Operations,” arguing that motivational taxonomies (state vs. commercial vs. social) are secondary to a shared structural logic of coordinated concealment; and Rogers2026-cy proposes “post-truth spaces” as a companion concept for fact-checkers, mapping how fringe clusters gain influence through proximity to mainstream discourse rather than direct virality. Adjacent applications of the CLSB lineage—Marino2023-9137f448 on Alternative Influence Networks and Italian COVID discourse, Song2025-yh on pro/anti-vaccine CLSB in the UK/US—demonstrate the method’s portability across health and political domains, while Song’s finding that CLSB is “ideologically agnostic” (serving both misinformation and credible public health messaging) complicates any assumption that coordination itself signals malfeasance.
Methodological pluralism and the limits of any single signature
A second cluster interrogates and extends the technical apparatus of detection itself, often in implicit dialogue with the CooRnet/CLSB tradition. Yang2025-iv responds directly to the vulnerability of timing-based signals (trivially manipulable) by proposing statistical regularities in sharing speed and frequency as more robust alternatives. Mannocci2025-ig and Iannucci2025-eg independently converge on the insight that coordination is inherently multimodal and temporal—no single action type (retweets, hashtags, URLs) suffices, and naive flattening across modalities loses structure; both advocate multiplex representations, with Iannucci adding a decay-weighted temporal kernel and Mannocci systematically comparing five operationalizations on UK election data. Minici2024-tf pushes toward generalization with a graph foundation model (IOHunter) designed to transfer across heterogeneous state-sponsored campaigns, while Gerard2025-br’s CANE/t-CANE framework reframes cross-platform influence structurally, via shared narrative-cluster membership rather than interaction traces, identifying a tiny population of “bridge users” who seed most cross-platform narrative migration between Truth Social and X. Domain-specific extensions push detection into new platform ecosystems: Luceri2025-tr into TikTok’s video-first environment, Rodriguez_Farres2025-sg into real-time bot detection on Bluesky, Kansaon2025-id into WhatsApp’s closed, encrypted groups in Brazil. Together these papers chart an arc from single-signal heuristics toward increasingly sophisticated, transferable, and platform-agnostic architectures—while Thiele2025-ol steps back to argue that the foundational category itself (Meta’s “coordinated inauthentic behavior”) conflates inauthenticity with manipulative intent, proposing instead a rational-choice typology oriented around attribution of hidden principals.
Skepticism, scale, and the question of impact
A countervailing thread within the corpus interrogates whether detected coordination actually matters. Di-Marco2025-aa formally models influence in retweet cascades and finds that real coordinated accounts in the 2019 UK election achieve influence far below optimal or even randomly-matched benchmarks—suggesting CIB may be structurally marginal rather than pivotal. Simeone2025-vo complements this from a different angle: examining Twitter’s deplatforming of Arizona election-audit accounts, the authors explicitly rule out CIB as an explanatory mechanism, showing that ripple effects stemmed from human hub/authority collapse, not bot coordination. The Meta-collaboration papers Appel2026-qr and its companion noauthor_undated-bm extend this skepticism to causal inference at scale: despite documenting massive reach for 2020 US election deceptive networks (37 million Facebook users), and showing that most of that reach is driven by reshares from non-network accounts rather than direct posting, the studies find that associations between exposure and downstream political outcomes largely evaporate once pre-exposure characteristics are controlled for—a sobering corrective to inflated causal narratives about influence-operation effects. Against this, Kim2026-wg and Bollenbacher2026-vz argue for measurable causal impact via more targeted designs: geo-racially targeted voter-suppression ads correlating with real turnout declines, and antivaccine tweet exposure causally linked (via epidemic modeling) to vaccine refusals and COVID deaths. The tension between these two pairs of studies—both methodologically rigorous, reaching opposite conclusions about causal traction—marks an unresolved fault line in the field: exposure and reach are not the same as effect, and the field is still calibrating how to move from behavioral detection to consequence.
Actors, infrastructures, and geopolitical variety
A substantial portion of the corpus supplies texture on who coordinates and why, moving from behavioral signature to political economy and organizational sociology. Poliakoff2026-fa’s OSINT analysis of IRA employee CVs revises the standard narrative of the “troll factory,” showing it recruited young, inexperienced graduates and was likely oriented toward Crimea rather than the 2016 US election—reframing the IRA as an ordinary media-labor-market actor rather than an exceptional covert unit. Gaw2025-ru extends this organizational lens to the 2022 Philippine elections, theorizing influence operations as political brokerage mediating between clients, platforms, and voters. Jovanovic-Harrington2026-ze documents Serbia’s SNS party mimicking grassroots authenticity in ways that evade standard bot-farm detection, while Kuznetsova2025-nu traces coordinated amplification among pro-government Telegram channels in Russia and Belarus, and Kulichkina2026-zk examines dual-use coordination (protest mobilization vs. state-aligned repression) during China’s COVID protests. Kim2026-br’s twenty-year study of Korean troll comments innovates methodologically (explainable, span-level rationale classification) while substantively showing that condemnation—not praise—is the dominant and most-amplified rhetorical strategy, a finding that complicates simple “boosting vs. attacking” typologies. Bastos2025-ol adds a visual-forensics dimension, probing whether troll farms carry identifiable visual signatures by country of origin. Pante2025-pq applies renewed methodological rigor to re-examine—and cast doubt on—prior claims of inter-state IO collaboration, underscoring how sensitive attribution claims are to control-dataset design. Collectively, these papers insist that coordination is not a monolithic phenomenon but takes shape differently across authoritarian, hybrid, and electoral-democratic contexts, and that organizational/labor perspectives (Poliakoff, Gaw) are as revealing as network-structural ones.
Emerging frontiers: platform response, generative AI, and theoretical stock-taking
The corpus closes on two forward-looking concerns. First, platform governance and moderation efficacy: Donovan2025-ws historicizes “misinformation-at-scale” as a structural feature of engagement-driven business models, showing how platforms retreated from CIB enforcement after 2021 political backlash; Oprea2025-lf documents Meta’s continued failure to curb hyperactive-user amplification during the 2024 Romanian EP elections despite explicit policy prohibitions; and FitzGerald2025-nv and Graham2025-gp examine how manipulation campaigns persist by appropriating emerging events and exploiting the platform “infrastructure of truth” itself (e.g., IStandWithPutin). Second, generative AI represents the field’s newest and most urgent extension: Schroeder2026-im theorizes “malicious AI swarms” as a qualitative escalation beyond human-driven botnets, while Orlando2025-ul provides empirical grounding via generative agent-based modeling, showing that LLM agents spontaneously reproduce coordination signatures (narrative convergence, synchronized amplification) as operational awareness increases—with mere mutual awareness nearly matching explicit deliberation. These papers gesture at where CooRnet-style detection must go next: toward systems capable of registering coordination among synthetic, adaptive, non-human actors, a challenge vera.ai’s WP4 alert infrastructure is positioned to confront as the empirical and theoretical work surveyed here converges on the recognition that coordination detection, once a niche behavioral signature, has become a necessary lens across elections, health communication, platform policy, and now the frontier of agentic AI.