An Italian Laboratory for Digital Political Communication

The backbone of this topic is a long-running empirical programme — referred to internally as MINE — that has tracked Italian electoral communication across more than a decade of platform change. Its earliest instalment, Giglietto2012-32b6b551, already poses the question that will recur throughout: can social-platform metrics stand in for electoral support? Testing Facebook Likes against the 2011 municipal elections, it finds no reliable correlation between popularity share and vote share, but a suggestive relationship between Facebook visibility and viability (the most-liked candidate wins or places second in 82% of races). This methodologically cautious beginning — mixed findings, explicit confirmation of contradictory prior literature — sets a tone of empirical skepticism that persists across the programme: metrics of attention are not simply proxies for political support, and the gap between visibility and influence must be measured, not assumed.

Iannelli2015-e0818c3e extends the inquiry to the “second screen,” showing how Twitter conversation around the 2013 general election talk shows remained tethered to broadcast logic rather than generating autonomous political agenda-setting. Its finding that participation is narrow and dominated by opinion/commentary rather than issue-framing anticipates a theme that later papers in the programme will formalize quantitatively: that attention online is unevenly distributed and strategically exploitable rather than organically democratic.

Partisanship, Insularity, and Coordinated Manipulation

By the 2018 general election and 2019 European election, the programme’s attention shifts from forecasting and participation to the structural dynamics of partisan news ecosystems. Giglietto2019-882f1900 introduces the concept of source “insularity,” showing that populist-aligned news sources (especially those favoured by the Five Star Movement) are consumed in more closed communities, and that M5S supporters strategically deploy shares to amplify favourable coverage while mobilizing comments to contest unfavourable coverage — a distinctive “cyber party” pattern not replicated by PD, Forza Italia, or the League.

This interest in strategic, collective online behaviour crystallizes methodologically in Giglietto2020-9d8acdd7, which proposes “coordinated link sharing behaviour” as an ecological, action-based alternative to content- or actor-centred disinformation detection — a move that becomes foundational for the rest of the programme. The approach is operationalized at scale in Giglietto2023-fa71a001, which builds a continuously updating workflow for surfacing coordinated accounts during the 2022 snap election, uncovering not only partisan hyperpartisan networks around M5S but adjacent commercial and religious-proselytism operations that exploit the same coordination infrastructure. Marino2023-9137f448 applies this lineage to a non-electoral but adjacent crisis — COVID-19 misinformation — tracing how an “Alternative Influence Network,” anchored by legacy-media remediation of expert statements, amplifies covid-skeptic narratives, demonstrating that the coordination-detection toolkit travels well beyond election periods narrowly defined.

Parallel work generalizes the rhetorical dimension of this manipulation. Mosca2026-yh reframes “fake news” accusations themselves as strategic weapons of negative campaigning rather than symptoms of misinformation, proposing a Strategic Delegitimization and Selective Amplification model for how Italian newspapers differentially amplify such accusations — a rhetorical-strategic companion to the behavioural coordination studies. Outside Italy, Oprea2025-lf documents a structurally similar phenomenon of hyperactive-user amplification on Facebook during Romania’s 2024 European Parliament campaign, underscoring that coordinated inauthentic sharing is not an Italian idiosyncrasy but a recurring vulnerability of platform architectures across European election cycles — and that platform self-regulation (Meta’s Community Standards) remains insufficiently enforced.

Methodological Infrastructure: From Clustering to LLMs

A distinct cluster of papers addresses how the Italian programme’s empirical claims should be computationally produced and validated. Giglietto2024-cbeb3f70 benchmarks embedding models for clustering Italian political news shared on Facebook before the 2018 and 2022 elections, finding that OpenAI’s text-embedding-3-large consistently outperforms the Italian-specific UmBERTo model — a result with direct bearing on how narrative/topic clusters are constructed in subsequent analyses. Marino2024-2fbc690f builds on this to articulate the validation challenges of a full “LLMs-in-the-loop” pipeline (classification, embedding, cluster labelling) applied to the same 2018/2022 datasets, arguing that genuinely novel validation protocols — not crowdsourced ones — are needed once LLMs outperform typical human annotators.

Paci2025-ag pushes this methodological scrutiny toward the limits of LLM competence, testing whether multilingual LLMs can interpret implicatures and presuppositions in the IMPAQTS corpus of Italian political speech; even the best model falls well short of expert human interpretation, suggesting that while LLMs are increasingly embedded as research infrastructure, their pragmatic blind spots — especially around politically loaded indirection — require caution. Achmann-Denkler2026-lx, working on German Instagram data rather than Italian, extends the same evaluative logic to multimodal models, showing GPT-4o substantially outperforms specialized computer-vision pipelines for visual political communication, and thereby offering a methodological parallel (and implicit validation route) for the kind of multimodal extensions the Italian programme might pursue next.

Platform Governance, Visibility, and Data Access

A more recent strand turns from describing political communication to interrogating the platforms that structure its visibility. Giglietto2025-1765bb4f uses the Meta Content Library to show that Meta’s political-content-reduction policy suppressed the Facebook reach of Italian parliamentarians well before its officially announced global rollout, with asymmetric effects that let extremist accounts offset per-post losses through sharply increased posting volume — a direct empirical rebuttal of Meta’s own transparency claims, read through the lens of algorithmic visibility as democratic infrastructure. Philipp2026-tl zooms out to the regulatory context enabling such research, surveying how the EU Digital Services Act has reshaped (and unevenly reopened) platform data access for election researchers, implicitly explaining why Meta Content Library-based work like Giglietto2025-1765bb4f has only recently become possible, and flagging the equity and reproducibility gaps that remain.

Conceptually, this governance-and-measurement concern was anticipated earlier in the programme by Giglietto2020-6278a4aa, which identifies three vectors by which big data has transformed political communication practice and scholarship: blurred organizational boundaries via networked publics, opaque micro-targeted advertising, and hard-to-measure strategic amplification. Read alongside the later empirical papers, this piece functions as the programme’s theoretical hinge — explaining why coordination detection, platform policy audits, and LLM-based pipelines all became necessary methodological responses to a media environment no longer legible through self-reported exposure or simple engagement counts.

Comparative Horizons: Populism, Polarization, and Misinformation Beyond Italy

The Italian case studies are embedded in, and implicitly tested against, a wider comparative literature on digital political communication’s pathologies. Eady2025-vm offers a general measurement framework for mapping the ideology of media, politicians, and publics in the US from URL-sharing behaviour, with findings — polarized sharing concentrated among ideologically extreme legislators, amplified under reduced electoral competition — that resonate with the insularity dynamics documented for Italian populist parties in Giglietto2019-882f1900. Guess2019-ym complicates the prevailing panic over “fake news” by showing that, in the 2016 US context, actual sharing of fake-news domains was rare and concentrated among older and more conservative users, a cautionary finding against overstating misinformation’s reach that pairs instructively with Mosca2026-yh’s argument that fake-news discourse is often more rhetorical weapon than empirical phenomenon.

Gaber2022-bk theorizes “strategic lying” in the Brexit and 2019 UK election context as an evolution of political spin rather than a post-truth rupture, directly relevant to how coordinated and rhetorical manipulation (as studied in the Italian coordination and SDSA papers) should be theorized — not as novel pathologies but as digitally amplified versions of established persuasion tactics. Tai2026-qk narrows this question to elite behaviour, examining how institutional role and ideology condition US elected officials’ own misinformation sharing, a question with clear resonance for — but not yet answered for — the Italian parliamentary context mapped in Giglietto2025-1765bb4f.

Cross-national platform studies further widen the frame: Darius2026-xl finds that radical populist parties enjoy disproportionate audience support (though not disproportionate posting activity) across all 27 EU states, most dramatically on TikTok, echoing the asymmetric, audience-driven amplification of M5S content found in Giglietto2019-882f1900 and the extremist-account resilience documented in Giglietto2025-1765bb4f. Balluff2026-bv demonstrates a different capture mechanism — government-advertising-driven media bias in Austria’s “Inseratenaffäre” — extending the political-parallelism concerns at the heart of the Italian programme (polarized-pluralist media systems, political-media entanglement) to another European democratic-corporatist case, using a similarly quasi-experimental computational design. Finally, the two Norwegian studies, Larsson2026-ro and Larsson2026-nl, trace a decade of party Facebook sentiment and engagement, finding that negativity drives shares and comments but not likes, and that incumbency — more than ideology — predicts negative campaigning; together they offer a longitudinal, multi-party counterpoint to the more episodic, single-election focus of the Italian papers, suggesting a natural next step for the MINE programme: extending its coordination- and visibility-centred toolkit into sustained, cross-electoral-cycle comparison.