The Shape of the Field: Two Founding Debates
Two long-running debates anchor everything filed here. The first is whether digital media segregate audiences into ideologically homogeneous enclaves. Early behavioral work — Bakshy2015-rn, Flaxman2016-lm, Barbera2015-fw, Messing2014-jc — found modest, topic-dependent segregation and argued that individual choice, not algorithms, does most of the sorting. A decade later, population-scale Facebook data revisits this with sharper tools: Gonzalez-Bailon2023-uy shows segregation is far higher than web-browsing studies suggested and asymmetrically concentrated in a homogeneously conservative corner of the ecosystem, while Nyhan2023-gb and the companion algorithm/reshare experiments (Guess2023-ur, Guess2023-ai) find that even when exposure to like-minded content is experimentally reduced, downstream attitudes barely move — a striking dissonance between exposure and effect that recurs across this whole corpus. Green2025-ap reframes the entire debate: source-level “domain” measures of partisanship mask enormous story-level heterogeneity, meaning much of the echo-chamber/moderate-media literature has been measuring the wrong unit of analysis, a methodological point sharpened further by Gaisbauer2025-by’s call for multi-level (story/outlet/content) cartographies of circulation and by Hartmann2025-px’s systematic review showing that disagreement about whether echo chambers “exist” is substantially an artifact of inconsistent operationalization. Barbera2015-je supplies the methodological backbone for much of this work — ideal-point estimation from network structure — later extended to video content by Lai2024-to and to LLM-based ideology inference by Lee2026-je, which shows political alignment leaking even from ostensibly nonpolitical talk. Schulz2026-ts closes this arc with a reframing: much apparent polarization online is false polarization produced by the self-selection of ideologically extreme users into posting, not a true reflection of population attitudes — a genuinely different mechanism from algorithmic curation.
The second debate is whether exposure to the other side helps or hurts. Bail2018-fk’s bot-follower experiment famously found backfire effects concentrated among conservatives; Dehghan2026-sy extends the pessimism to Reddit, showing that even discursively similar subreddits share almost no users, sources, or cross-commentary — polarization there is “sedimented” as a precondition of platform architecture and moderation, not merely discourse. Against this, Esau2025-tf proposes deliberative reciprocity and inclusive listening as normative correctives, and van-Eck2026-xg’s review of the depolarization literature finds that such correctives are rarely justified by the evidence base scholars themselves produce — a sobering meta-commentary on the whole field’s practical recommendations.
Do Platforms and Algorithms Cause Polarization?
A cluster of large field experiments interrogates the algorithmic-amplification hypothesis directly. The Meta 2020 election studies (Guess2023-ai, Guess2023-ur, Nyhan2023-gb, Gonzalez-Bailon2023-uy) collectively produced a puzzle: algorithms and reshares dramatically reshape what people see, yet produce almost no measurable movement in polarization, knowledge, or turnout. Gauthier2026-iq partially resolves this puzzle for X/Twitter: switching users onto the algorithmic “For You” feed does shift attitudes rightward, but the effect is asymmetric (it doesn’t reverse when switched off) and works through a persistence mechanism — algorithmic exposure changes who users follow, and those follows outlast the algorithm itself. This account of “sticky” behavioral change rather than direct persuasion helps explain why the earlier Meta null results and Gauthier’s positive results are not actually contradictory. Brady2026-ln’s Bluesky Registered Report pushes the mechanism further: engagement-based feeds amplify intergroup, moralized, emotional content and distort perceptions of political-dialogue norms rather than directly reshaping attitudes — pointing toward a norm-misperception pathway as the psychologically real channel connecting algorithms to animosity. Wan2026-ai examines the demand side of this same pipeline, showing political efficacy conditions how people select political content from algorithmic feeds like Google Discover. Complicating platform-level generalizations further, Mosleh2024-op shows engagement with partisan and low-quality news diverges substantially across seven platforms, cautioning against Twitter- or Facebook-centric conclusions — a caution Bouchafra2026-ts and Holland_Levin2026-qx make concrete by showing how the same political content (Sweden Democrats’ securitizing memes; the politicization of “woke”) is deployed with platform-specific rhetorical logics across Facebook, X, TikTok, and YouTube.
Governance interventions receive similarly mixed verdicts. Bouchaud2026-np argues X’s Community Notes bridging algorithm structurally undermoderates precisely the polarizing content most in need of correction, since such content is least likely to attract cross-partisan agreement — a finding with direct implications for the global rollout of bridging-based moderation. Votta2025-xz and Inacio-da-Silva2026-zf extend platform-accountability concerns to the opaque economics and irregularities of political advertising delivery.
Affective Polarization as Identity, and Its Elite Drivers
Iyengar2019-jj remains the theoretical keystone distinguishing affective polarization (partisan animus as social identity) from ideological/issue polarization, and tracing its spillover into economic and social life. Osmundsen2021-et operationalizes this for misinformation sharing, showing fake-news dissemination is best explained by out-party hatred rather than ignorance or trolling — fake news is simply the extreme tail of ordinary partisan sharing. Van_Erkel2026-mk extends the identity logic one step further with the “hostile misinformation effect”: citizens across Germany, the Netherlands, and Poland believe their own side is disproportionately targeted by misinformation, a bias amplified by interest, identity strength, and (especially) right-wing orientation. Schemer2026-mh’s specification-curve analysis shows that even the magnitude (not direction) of the partisan-media/polarization relationship is highly sensitive to how media slant and polarization are measured — reinforcing the methodological caution running throughout this literature. Grusauskaite2026-po offers a complementary, interactionist account of how affective boundaries are built discursively: “Great Reset” communities across six platforms sustain a shared oppositional identity through symbolic boundary work and reclaimed stigma rather than through platform affordances alone, echoing Elfes2026-jb’s argument that polarization operates at the level of narrative — divergent actantial roles assigned to key actors — as much as at the level of opinion or interaction.
At the elite level, Knupfer2025-vt’s “logic of connective faction” and Bennett2025-xs’s “digital surrogate organizations” framework jointly argue that radicalization is an organizational phenomenon: hyper-partisan media and digitally networked factions provide measurable incentives (engagement, visibility) that reward ideological extremity within parties, restructuring intra-party power rather than simply informing or misinforming voters. Tornberg2025-ir’s 26-country, 32-million-tweet analysis gives this claim comparative teeth: neither populism nor right-wing ideology alone predicts elite misinformation sharing, but their conjunction — radical-right populism — is the strongest predictor, tying disinformation to a specific ideological formation rather than a generic media pathology. Benkler2018-lw and Bennett2018-ys supply the foundational architecture for this claim, arguing the American “epistemic crisis” is asymmetric and institutional rather than technologically determined or symmetrically bipartisan, a diagnosis Freelon2020-yp situates within the broader “disinformation order” and definitional history of the field.
The Brazilian and Latin American Front
A dense empirical cluster examines Brazil as a paradigmatic Global South case of hybrid-media polarization. Rodarte2026-dk documents how Brazilian parliamentarians contested epistemic authority during the Manaus COVID crisis, with pro-Bolsonaro legislators severing ties to mainstream media entirely while opposition figures mimicked journalistic form to borrow credibility. Marino2026-he offers the deepest longitudinal window into the pro-Bolsonaro Facebook ecosystem itself, finding — against the assumption of stable echo chambers — that affective engagement in this hyperpartisan network was highly volatile, with a marked de-escalation of emotional intensity around Lula’s inauguration and the January 2023 coup attempt. Kansaon2025-id traces the WhatsApp infrastructure connecting online coordination to offline mobilization, documenting how coordinated accounts drove misinformation-fueled doxxing of Supreme Court justices and the January 8 uprising; Ventura2025-sw complements this with an experimental WhatsApp-deactivation design probing the platform’s distinct causal role. Rossini2026-jn and Rossini2026-mj together show how belief in electoral misinformation eroded both political tolerance and trust in the 2022 election results — with mainstream news use acting as a democratic-resilience counterweight even among Bolsonaro voters — while Ventura2026-yc documents a striking coda: the 2024 Brazilian X ban produced a durable rightward “sorting ratchet” in the platform’s news environment, since conservative users disproportionately circumvented the ban while liberal users exited. Read together, this cluster substantiates the topic description’s emphasis on pro-Bolsonaro networks and cross-national Latin American comparison, while consistently emphasizing platform interdependence (Facebook, WhatsApp, X) over any single-platform story. Gaw2025-ru’s Philippine case extends the Global South lens beyond Brazil, reconceptualizing influence operations as political brokerage infrastructure rather than isolated propaganda.
Radical-Right Visibility and the Legacy-Media Interface
A parallel strand asks not how citizens polarize but how legacy and social media jointly amplify the radical right’s visibility. Iris2026-pg finds systematic overrepresentation of Radical Right actors in mainstream European news coverage ahead of the 2024 EU elections — even in countries, like Ireland, where such parties hold no seats — undercutting radical-right claims of media marginalization. Darius2026-xl shows the mirror pattern on social platforms: populist radical right (and far-left populist) parties enjoy disproportionate audience engagement rather than merely posting more, an asymmetry most extreme on TikTok. Bouchafra2026-ts and Much2026-gu each detail platform-specific mechanisms of normalization — securitizing memes targeting young voters, and “gendered media spaces” in entertainment podcasts that expose young men to incidental conservative content — while Gattermann2025-yx and Karlsson2026-hd examine downstream public reactions, the latter showing that media narratives about digital election interference themselves produce asymmetric partisan perceptions of societal division. Yoo2026-ev traces how “fake news” as a rhetorical weapon is deployed symmetrically by left and right hyperpartisan outlets to redraw Hallin’s spheres of legitimacy and deviance, complementing Balluff2026-bv’s demonstration of covert media capture (Austria’s “Inseratenaffäre”) as another channel by which political actors bend legacy media coverage in their favor.
Legacy Frameworks, Comparative Reach, and Method
Underpinning much of this newer work is Chadwick2013-ns and Chadwick2015-ii’s hybrid media system theory, which insists that old and new media logics interpenetrate rather than one displacing the other — a framework explicitly invoked by Rodarte2026-dk, Knupfer2025-vt, and Iris2026-pg. Humprecht2020-gd operationalizes comparative resilience to disinformation across eighteen Western democracies, finding the United States a structurally distinct, unusually vulnerable case — a finding Falkenberg2026-ka uses to indict the field’s persistent US-centrism and call for globally equitable polarization research, a call this bibliography’s Brazilian and Philippine cases begin to answer. Gaber2022-bk and Berlinski2023-rj extend the elite-strategy lens to the UK and US respectively, showing “strategic lying” and unsubstantiated fraud claims erode electoral trust largely irrespective of fact-checking corrections. Methodologically, the corpus increasingly turns to LLMs and unsupervised pipelines to detect these dynamics at scale — Sarmiento2025-as, Elfes2026-jb, Lee2026-je, Eady2025-vm — while Anwar2024-34dba628, Larsson2026-ro, and Kalsnes2025-zb converge on a consistent affective-engagement finding across Facebook Reactions studies: anger and negativity reliably outperform positivity in driving shares, a mechanism Zhao2025-ny and Arceneaux2026-xk link explicitly to bot-driven and hashtag-hijacking amplification tactics. Together with the science-communication cases of Ghezzi2023-8bebc91f, Lieu2025-nl, and Giglietto2019-882f1900/Giglietto2017-4375de2f (Italian populist news insularity; Charlie Hebdo dissent), and the narrative-sensemaking work of Prochaska2025-ef and Suau_Martinez2026-lv on election-fraud “deep stories” and recurring disinformation narratives, this final cluster underscores the volume’s central methodological throughline: polarization research is only as trustworthy as the conceptual and measurement choices — echo chamber operationalization, ideology scaling, media slant scoring — that variously reveal or manufacture it, a point Kim2026-wg, Tai2026-qk, and Kristensen2025-ni each reinforce from the angle of targeted suppression, elite accountability, and platform-specific discursive effects, respectively. Makeev2026-ma closes the loop by reminding us that even authoritarian propaganda bifurcates strategically by audience — a structural analog to the domestic partisan sorting this entire literature otherwise documents.