Framing the field: from single-country alarm to comparative method
The papers gathered here trace an arc from early, often US-centric alarm about “fake news” toward a mature comparative science of how social media infrastructures condition election campaigns across very different political systems. Chadwick2015-ii supplies the foundational theoretical vocabulary — the “hybrid media system” in which older and newer media logics interact rather than one displacing the other — that almost every later empirical study in this collection implicitly operationalizes, whether describing Italian talk-show second-screening in Iannelli2015-e0818c3e or Brazilian parliamentarians competing with journalists for epistemic authority in Rodarte2026-dk. Humprecht2020-gd pushes this comparative instinct further by modeling disinformation resilience as a structural, cross-national property of media and political systems, explicitly warning that findings from the US and UK should not be generalized — a caution the rest of the corpus takes seriously by multiplying national cases: Serbia (Jovanovic-Harrington2026-ze), the Philippines (Gaw2025-ru), Austria (Balluff2026-bv), Scandinavia (Kalsnes2025-zb, Larsson2026-ro), Romania (Oprea2025-lf), and the UK (Gaber2022-bk, Karlsson2026-hd).
The Italian laboratory: a decade of detecting coordination
The Italian case functions as a methodological proving ground across the 2012, 2018, 2019, and 2022 elections. Giglietto2012-32b6b551 opens the series with a cautionary note on Facebook “Likes” as an electoral predictor, finding the metric noisy but not meaningless — a finding that anticipates later arguments that platform signals require careful, actor-level interpretation rather than naive forecasting. Giglietto2019-882f1900 then documents how M5S and Lega supporters cultivate more “insular” Twitter news ecologies, with the Five Star Movement’s cyber-party apparatus strategically modulating shares versus comments to amplify favorable coverage — a template for populist attention management echoed later in Darius2026-xl and Tornberg2025-ir. The pivotal methodological move comes with Giglietto2020-9d8acdd7, which reframes disinformation detection around “coordinated link sharing behavior” rather than content veracity or bad-actor lists, a reframing extended into real-time monitoring by Giglietto2023-fa71a001, whose workflow surfaces not only hyperpartisan M5S networks but also clickbait and religious-proselytism operations hiding behind ostensibly apolitical pages. This lineage culminates in explicitly computational self-reflection: Giglietto2024-cbeb3f70 and Marino2024-2fbc690f turn the Italian corpus into a testbed for validating LLM-based embeddings and “LLMs-in-the-loop” pipelines, raising the methodological stakes from detecting coordination to trusting automated interpretation itself — a concern that reappears in Mosca2026-yh’s argument that “fake news” accusations in the Italian press are themselves strategic rhetorical weapons rather than neutral descriptions of information quality.
This Italian emphasis on coordination as a behavioral signature, independent of bot-like automation, is directly generalized by Nizzoli2020-cf’s UK 2019 study, which shows coordination and automation to be orthogonal and that the most tightly coordinated communities are small activist fringes rather than mainstream parties, and by Keller2019-nk’s ground-truthed South Korean case, which demonstrates that even well-resourced, human-staffed astroturfing campaigns leave detectable behavioral traces yet achieve only modest discursive impact. Jovanovic-Harrington2026-ze extends the coordination problem into authoritarian terrain, arguing that Serbia’s SNS has evolved beyond centralized bot farms toward more mimetic, harder-to-detect grassroots simulation, while Oprea2025-lf’s Romanian audit of “hyperactive users” on Facebook shows the same mimicry operating inside an EU electoral context nominally governed by the Digital Services Act.
Brazil’s post-Bolsonaro digital battlefield
A second major cluster follows Brazil’s uniquely polarized, WhatsApp-centric information ecosystem from the 2018 ad-auditing experiment in Inacio-da-Silva2026-zf through the Bolsonaro-era crises. Rodarte2026-dk shows how during the Manaus health collapse, government critics, Bolsonaro allies, and local brokers each pursued distinct strategies of epistemic authority, fragmenting the very possibility of a shared crisis narrative. This fragmentation hardens into electoral consequence in Rossini2026-jn and Rossini2026-mj, companion panel studies showing that belief in electoral fraud misinformation — often channeled through messaging apps — both erodes political tolerance and undermines trust in election results, even as mainstream news use (even among Bolsonaro voters) bolsters democratic resilience. Kansaon2025-id traces how this belief ecosystem materializes offline, documenting coordinated WhatsApp networks that doxxed Supreme Court justices and helped mobilize the January 8, 2023 insurrection, while Ventura2025-sw’s WhatsApp deactivation experiment attempts to causally isolate the platform’s distinct contribution to these dynamics. Marino2026-he’s three-year longitudinal study of pro-Bolsonaro Facebook communities complicates the “stable echo chamber” assumption underlying much of this work, showing instead that affective engagement destabilizes sharply around events like Lula’s inauguration and the coup attempt. The arc closes, post-Bolsonaro, with Ventura2026-yc’s study of Brazil’s 2024 X ban, which shows how a state-imposed platform sanction intended to curb misinformation instead produced a durable rightward “sorting ratchet” — a finding that complicates any easy regulatory solution to the dynamics documented throughout this cluster.
Political advertising, platform governance, and the data-access bottleneck
A parallel strand interrogates political advertising and the infrastructures that make (or prevent) its study. US-focused work establishes both the promise and the limits of causal inference here: Kim2026-wg finds that geo-racially targeted voter-suppression ads in 2016 measurably depressed turnout among targeted subpopulations, while the large Meta-collaboration studies Appel2026-qr, Allcott2025-jb, and Gonzalez-Bailon2024-rq complicate any simple story of advertising-driven harm — deceptive networks reach millions mostly via unaffiliated reshares, and experimentally removing political ads from feeds produces no detectable attitudinal effects, even as misinformation exhibits distinctive viral diffusion patterns. Votta2025-xz turns attention to the delivery algorithms themselves, arguing that opaque ad-ranking systems, not just advertiser targeting, shape reach and cost.
This advertising research is inseparable from a growing crisis of data access, which several papers treat as a research object in its own right. Philipp2026-tl surveys how the EU’s Digital Services Act has reopened — unevenly — platform APIs for election researchers, a theme sharpened by Jurg2025-ur’s critical audit of YouTube’s moderation and “authoritative source” labeling during the 2024 EP elections, which finds publisher-context labels applied inconsistently across languages and broadcasters. Lukito2026-nb synthesizes this landscape into a “price, proficiency, or permission” framework across 72 data-access tools, documenting a severe imbalance toward production over consumption data and a WEIRD-centric infrastructure that Schulte2026-df’s German 2025 case study illustrates concretely, showing how X’s rate limits and TikTok’s algorithmic filtering introduce systematic blind spots into even well-resourced multi-platform research.
Computational methods and the German case
The German federal election supplies this topic’s main methodological showcase. Achmann-Denkler2026-lx validates multimodal LLMs (GPT-4o) against specialized computer-vision pipelines for analyzing visual political communication on Instagram during the 2021 campaign, finding that prompt-based approaches approach human-level annotation while revealing distinct “concentrated visibility” strategies between candidate and party accounts. Schulte2026-df, examining the 2025 election, extends this multimodal logic across platforms, showing that TikTok’s textual, vocal, and facial channels operate nearly independently — evidence for a “politainment” register of critical content delivered in calm tones. Both papers sit alongside Le-Mens2025-qz’s more general demonstration that LLMs can reliably position political texts on ideological scales by “asking and averaging,” suggesting these computational advances are rapidly becoming infrastructure for comparative campaign research generally, not merely German or Italian case studies.
Comparative Europe: populism, visibility, and the radical right
A cluster of 2024 European Parliament election studies builds directly on the Italian and German methodological groundwork to ask whether digital campaigning structurally advantages the radical right. Darius2026-xl’s cross-platform analysis of 401 parties across all 27 member states finds that populist radical right parties enjoy a pronounced support advantage (especially on TikTok) that exceeds any advantage in posting activity, a pattern echoed in Bouchafra2026-ts’s close reading of the Sweden Democrats’ platform-tailored securitizing visual rhetoric and in Perez-Curiel2026-ld’s account of AI-generated multimedia in far-right Instagram and X campaigning. Iris2026-pg complements these supply-side findings with a demand-side one: radical right actors receive disproportionate news visibility relative to polling and actual results across several countries, intensifying in the campaign’s final weeks regardless of outlet orientation. Tornberg2025-ir’s 26-country parliamentary Twitter analysis supplies the elite-behavior mechanism underlying these visibility patterns, showing that neither populism nor right-wing ideology alone predicts misinformation sharing — only their combination, radical-right populism, does — while Gattermann2025-yx shows that far-right electoral performance itself shapes voters’ disinformation concerns, closing a feedback loop between elite strategy, media visibility, and public perception that Humprecht2020-gd’s structural resilience framework anticipated.
The US baseline: concentration, selective exposure, and the limits of influence
Although geographically outside the topic’s core cases, a substantial US literature functions as this collection’s methodological and empirical baseline, against which the German, Brazilian, and Italian findings are implicitly measured. Barbera2015-je establishes the ideal-point methodology later echoed in Eady2025-vm’s URL-sharing ideology model, both showing that ordinary politically engaged users, not elites, drive the most polarized content. A consistent finding across Guess2020-rr, Grinberg2019-ua, Guess2019-ym, and Eady2023-xg is that exposure to fake news and foreign influence content, while real, is heavily concentrated among small, already-committed, often older and more conservative subpopulations, and dwarfed by mainstream media exposure — a pattern of concentration that resurfaces internationally in Appel2026-qr’s deceptive-network reach figures and in Nenno2025-xa’s 24-country comparison of news values in flagged misinformation, where the US and Brazil emerge as distinctive high-polarization outliers. Brady2026-ln’s Bluesky field experiment pushes this literature toward intervention, showing that engagement-based ranking algorithms distort perceived social norms around political dialogue and that a “diversified extremity” algorithm can partially correct this without sacrificing platform enjoyment. Much2026-gu and Lukito2026-il extend the US frame to podcasts and cross-platform curation on Truth Social, while Starbird2025-jj and Prochaska2026-pr develop complementary frameworks — evidence-frame dynamics and recurring narrative “tropes” — for understanding how election-integrity rumors are assembled discursively rather than simply fabricated. Karlsson2026-hd’s UK survey experiment closes this cluster by showing that even coverage of election interference, independent of its real effects, can asymmetrically deepen partisan perceptions of division.
Throughlines and open frontiers
Read together, these papers converge on several shared claims that justify treating them as one comparative conversation rather than isolated national literatures: that coordination, not content or automation per se, is the most durable unit of analysis for manipulation (Giglietto2020-9d8acdd7, Nizzoli2020-cf, Keller2019-nk); that misinformation’s political consequences are better explained by elite strategy and institutional context (radical-right populism, media capture, “strategic lying”) than by simple exposure volume (Tornberg2025-ir, Balluff2026-bv, Gaber2022-bk, Knupfer2025-vt); and that the very capacity to study these phenomena comparatively is now bottlenecked by platform governance and data access as much as by theory (Lukito2026-nb, Philipp2026-tl, Jurg2025-ur). The newest wave — computational pipelines built on LLMs for classification, embedding, and visual analysis (Marino2024-2fbc690f, Giglietto2024-cbeb3f70, Achmann-Denkler2026-lx, Le-Mens2025-qz) — promises to make truly comparative, multi-platform, multilingual election research newly feasible, but also imports fresh validation problems that this literature, especially its Italian and German strands, is only beginning to confront systematically. The Brazilian cases, meanwhile, remind the field that platform effects are not symmetric across the Global North and South: WhatsApp’s closed architecture, state-level platform bans, and weak-institution contexts generate dynamics — real-world mobilization, durable partisan sorting after deplatforming — that the US- and Europe-centric theoretical apparatus inherited from Chadwick2015-ii and Humprecht2020-gd does not yet fully capture.