Origins: mapping the Italian hybrid media ecology (2013–2019)
The program’s earliest work treats Italy as a strategic case for studying how television, Twitter, and Facebook co-produce political attention outside the Anglo-American context that dominates the field. Iannelli2015-e0818c3e establishes the baseline logic — a hybrid-media reading of the 2013 election in which Twitter functions as a second screen to talk-show television, with narrow, TV-driven participation rather than autonomous agenda-building. This hybridity framing recurs, refined, in Giglietto2019-882f1900, which shifts the object of analysis from television to news sources circulating on Facebook and Twitter during the 2018 campaign. Its central discovery — that populist-aligned sources (M5S above all) are more “insular” on Twitter, and that insularity predicts a Facebook engagement signature of amplification-versus-contestation — becomes a template the program returns to repeatedly: an audience’s structural position in the media ecology, not just its partisanship, shapes how it engages with news.
From ecology to coordination: detecting inauthentic collective action
The decisive methodological turn is Giglietto2020-9d8acdd7, which reframes disinformation research around coordinated link sharing behavior rather than content veracity or bad-actor lists, using the 2018 and 2019 Italian campaigns to show that coordinated networks disproportionately spread problematic domains and include previously flagged disinformation sources. This ecological, behavior-first logic becomes the program’s methodological signature and is operationalized at scale in Giglietto2023-fa71a001, which builds an iterative workflow to keep coordination lists current across the 2022 snap election, surfacing not only hyperpartisan political networks but click-economy and religious-proselytism operations that content-based detection would miss. Oprea2025-lf extends this actor-based lens geographically (2024 Romanian EP campaign), documenting hyperactive-user amplification on party pages and reinforcing the program’s recurring argument that platforms’ stated policies against inauthentic behavior are not matched by enforcement.
Explaining capture and rhetoric: why coordination and coverage bias occur
A parallel strand asks why such distortions arise structurally rather than only how to detect them. Balluff2026-bv documents Austrian “journalism for sale” via government advertising, showing media capture mechanisms operating even in consolidated democratic-corporatist systems — a comparative counterpoint to Italy’s own polarized-pluralist media system. Mosca2026-yh turns the lens on rhetoric itself, proposing a Strategic Delegitimization and Selective Amplification model in which “fake news” accusations in the Italian press function as negative-campaigning weapons rather than epistemic claims — reframing disinformation studies from information quality to political strategy, a natural complement to the coordination-detection papers’ behavioral focus. Ducci2022-10cb5d70 rounds out this account of gatekeeping and amplification by showing, in the health domain, how Google News Italia’s aggregation choices and Facebook engagement dynamics jointly shape which stories reach publics — evidence that the mechanisms driving visibility distortions in electoral coverage (studied in Giglietto2019-882f1900 and Balluff2026-bv) generalize beyond politics narrowly defined.
Methodological maturation: from BERT to LLM-in-the-loop pipelines
As the empirical objects grow (millions of posts across 2018 and 2022), the program’s tooling shifts from bespoke network algorithms to language-model-driven pipelines. Giglietto2024-cbeb3f70 benchmarks embedding models for clustering Italian political news, finding that general-purpose LLM embeddings (text-embedding-3-large) beat Italian-specific BERT models (UmBERTo) for semantic coherence — a finding operationalized in Marino2024-2fbc690f, which documents the design and validation challenges of a full LLMs-in-the-loop pipeline (classification, embedding, labeling) applied to the same 2018/2022 Facebook corpora, arguing that validating such pipelines requires new, task-specific, expert-driven protocols rather than crowdsourced benchmarks. Paci2025-ag probes the limits of this LLM turn specifically for Italian political discourse, showing that even strong multimodal LLMs struggle to interpret implicature and presupposition in real political speech — a sobering complement to the more optimistic clustering results, underscoring that automation gains are task-dependent. Le-Mens2025-qz generalizes the LLM-as-measurement-instrument logic beyond Italy, proposing an ask-and-average method for ideological scaling that the program’s Italian pipelines implicitly anticipate.
The 2022 election and the platform-policy shock
Giglietto2025-1765bb4f represents a maturation of the program’s object of study: rather than detecting actor behavior, it uses the Meta Content Library (a DSA-enabled data source) to show that Meta’s political-content-reduction policy suppressed Italian MPs’ Facebook reach months before its announced global rollout, with extremist accounts compensating through posting volume — a finding that closes a loop opened by the earlier coordination papers, since visibility is now shown to be shaped as much by opaque platform policy as by coordinated manipulation. This paper marks the program’s pivot toward treating platform governance itself, not only user/actor behavior, as a research object.
Widening the aperture: comparative and infrastructural turns
The most recent cluster generalizes the Italian program’s questions to other elections and to the platforms’ own compliance infrastructures. Achmann-Denkler2026-lx and Bouchafra2026-ts extend the visual/multimodal analysis largely absent from the Italian corpus to German Instagram and Swedish TikTok/cross-platform campaigning, respectively. Kalsnes2025-zb and Arceneaux2026-xk extend the affective-engagement and bot-agenda-building questions raised implicitly in Giglietto2019-882f1900 to Scandinavian Facebook reactions and U.S. midterm bot activity. Iris2026-pg and Gattermann2025-yx scale the visibility-bias question to the 2024 European Parliament elections across five and multiple EU countries respectively, asking whether radical-right overrepresentation and disinformation concern are structural, cross-national phenomena rather than Italy-specific ones. Meanwhile Philipp2026-tl, Jurg2025-ur, McNally2025-dn, and Schulte2026-df collectively theorize the DSA-era data-access regime itself — surveying new APIs, auditing YouTube’s authoritative-source ranking and removal opacity, demonstrating that Facebook’s News Feed algorithm is empirically detectable rather than a “black box,” and arguing that multi-platform election research needs shared, reusable infrastructure rather than one-off scraping solutions. Together these papers reposition the Italian program’s founding methodological commitments — actor- and behavior-based detection, workflow reproducibility, platform accountability — as a template now being tested, defended, and extended across the wider European electoral-communication landscape.