From open APIs to regulated access: a shifting infrastructure
The papers gathered here narrate a single arc: the closure of the open social media data ecosystem that sustained two decades of digital research, and the uneven, contested attempt to rebuild access through regulation. Freelon2024-sc provides the canonical periodization — from API “prehistory” through laissez-faire openness, authentication, and post-Cambridge Analytica restriction — while Bastos2025-ya offers a retrospective eulogy to the Twitter APIs that anchored a distinct subfield of “Twitter studies,” and Murtfeldt2025-wu bibliometrically quantifies the loss: a 13% decline in Twitter-based publications in 2024 after the $42,000/month paywall. Yang2026-tq’s systematic review confirms the pattern at the level of the whole social sciences: empirical social media research tripled through the early 2020s before plateauing and declining as Twitter/X and Facebook APIs closed. Giglietto2026-855a54cb reads Twitter’s “Wild West to Walled Garden” trajectory as paradigmatic of a structural pattern visible also in CrowdTangle’s shutdown, Reddit’s pricing, and TikTok’s restrictions.
Article 40, the DSA, and the promise of regulated access
A second cluster examines whether the EU’s Digital Services Act can institutionalise what platform philanthropy could not. Ohme2026-nv, Pierri2025-hm, Philipp2026-tl, and de-Vreese2026-zx all treat Article 40 as a genuine paradigm shift — from discretionary access to a binding right — while documenting platform foot-dragging, opaque procedures, and resource asymmetries. De Vreese and Tromble in particular reframe current turbulence as evidence that regulation is finally biting rather than failing. Peters2026-mo shows how “data quality,” largely absent from the original DSA text, was inserted into the final Delegated Regulation through academic and NGO advocacy against platform resistance — with Meta strategically inverting the argument to claim that platform data is too poor to be worth sharing. Crosset2026-mq situates the DSA within a broader comparative typology of platform regulation, identifying it as a “flow optimisation” regime that governs circulation through bureaucratic audit rather than direct content control.
What the new access actually delivers
Empirical audits of the new infrastructure are sobering. Entrena-Serrano2025-gw documents persistent limitations in TikTok’s DSA-mandated Research API; Rieder2025-ju shows YouTube’s search API is “forgetful by design,” with findable videos decaying sharply after 20–60 days, making retrospective study of events like elections nearly impossible; Jurg2025-ur audits YouTube’s moderation transparency around the 2024 EU Parliamentary elections and finds inconsistent publisher labels and vague removal statements. Schulte2026-df reframes these problems as a research-infrastructure question, arguing that platform observability must be treated as reusable scaffolding rather than ad hoc collection. Lukito2026-nb’s taxonomy of 72 election-research tools crystallises the underlying inequity as a “price, proficiency, or permission” problem that systematically excludes Global South and early-career scholars.
Industry influence and the politics of platform-academic collaboration
A sharp parallel literature interrogates the collaborations that did materialise. Bak-Coleman2026-mk finds that roughly half of high-profile social media papers in Science/Nature/PNAS have undisclosed industry ties, with an estimated 80% “industrial saturation” once editors and reviewers are included; Bak-Coleman2025-pm generalises this into a tobacco-and-pharma analogy for tech research. Heiss2026-qv makes the structural case that platforms’ exclusive data control creates unprecedented dependency, and Munger2025-cz delivers a metascientific critique of the Meta2020 partnership, arguing that field experiments on rapidly-mutating platforms have low temporal validity and that bigger collaborations cannot, in principle, keep pace. Allen2025-ot takes the opposite tack, championing platform-independent experimentation via browser extensions and LLM reranking as a way out of the access bind.
Workarounds: donation, scraping, and platform-independent methods
Where official access fails, researchers improvise. Bouchaud2026-lr reconstructs X’s recommender embedding from 2.5 million donated “Who to Follow” recommendations and shows the system encodes users’ Left–Right ideology with ρ=0.887. Inacio-da-Silva2026-zf deploys a volunteer browser plugin to audit Brazilian electoral ads; Efstratiou2025-gs uses simultaneously-collected algorithmic and chronological feeds to isolate the Musk-era amplification of right-leaning and “agitating” content; Brady2026-ln uses Bluesky’s open architecture to run a registered field experiment on prosocial algorithm design — implicitly making the case that open infrastructure is itself a research methodology.
What Meta data still reveals about platform governance
A productive vein of work mines the surviving Meta Content Library and URL Shares Dataset to expose platform interventions Meta itself has not disclosed. Giglietto2025-1765bb4f detects, via structural breakpoint analysis, that Meta’s political-content reduction policy began in Italy ten months before its announced global rollout and cut MP reach by 72%; Giglietto2026-632ef967 shows that sharing-to-views amplification on Facebook fluctuates with governance interventions like the 2020 “break the glass” measures, contradicting the assumption that algorithmic curation is a stable background. McNally2025-dn uses CrowdTangle data on Guardian articles to show that News Feed changes produce measurable, lagged engagement effects roughly 19–24 days after rollout, arguing against the “black box” framing. Giglietto2022-b30e8b4e cautions that the 100-share threshold in Meta’s URL dataset itself produces artefactual cross-country similarity, while Rossi2023-847d5a9f uses the same dataset for European comparison. Hurcombe2025-cs reads Meta’s Newsroom discourse as a strategic frame that exports terms like “coordinated inauthentic behaviour” into regulation while excluding hyperpartisan verified accounts and Facebook Groups from view; De2026-ld generalises this as “changecraft” across Meta, TikTok, YouTube, and X.
Moderation, fact-checking, and the retreat from trust and safety
Several papers track the political-economic unwinding of platform governance commitments. Donovan2025-ws historicises the rise and retreat of trust-and-safety regimes between 2016 and 2021; Moran2025-qn documents the field’s destabilisation through interviews with practitioners; Cazzamatta2026-lo shows that Meta’s portrayal of fact-checkers as censors is empirically unfounded, with removal occurring in only ~30% of cases. Farkas2026-lr examines how European fact-checkers rhetorically defend platform partnerships even as they criticise them, and Renault2025-uh tests the X Community Notes alternative, finding that Republicans are flagged 2.3 times more often than Democrats even under a crowdsourced bridging algorithm — undermining the bias rationale for abandoning professional fact-checking. Tonneau2025-bv uses DSA-mandated transparency reports to document stark cross-linguistic inequities in moderator allocation, with Global South languages receiving as little as 7.5% of English’s per-content investment on X.
Algorithmic effects, deplatforming, and platform bans
Adjacent work measures what platform governance actually does to discourse. Efstratiou2025-gs shows the post-Musk algorithm rewards proximity to Elon Musk and “agitating” content; Rieder2026-pp argues YouTube’s recommendation system outpaces moderation, sustaining an “ambient” Tate-space after deplatforming; Karo2026-dn documents how jihadist content evades TikTok moderation through algorithmically legible vernaculars. Simeone2025-vo shows targeted Twitter deplatforming of election-audit accounts produced large network ripple effects beyond the banned nodes themselves. Ventura2026-yc finds that Brazil’s nationwide X ban produced a durable partisan “sorting ratchet,” and Vincent_undated-re tracks ongoing mis/disinformation prevalence under European monitoring regimes. Votta2025-xz examines how ad delivery algorithms shape political campaign reach, while Schiffrin_undated-gi extends the gatekeeper-liability argument to deepfake financial fraud.
Migration, post-social formations, and the moving research object
A final cluster asks whether the object of study itself is dissolving. Wang2026-ub documents the failed migration of academic Twitter to Mastodon; Bruns2026-yv writes Twitter’s obituary; Tornberg2026-lc argues that algorithmic broadcasting, semi-private micro-communities, and AI-mediated communication have fragmented the “social media” object that animated two decades of scholarship. Weinbrand2026-sf extends this concern to generative search, framing Google’s AI Overviews as a platform-infrastructural redefinition of online knowledge. Mahl2026-hc’s Delphi study of expert opinion rates platform governance and journalism interventions as the most important responses to misinformation, while Lewandowsky2026-ob argues bluntly that platforms must be held accountable for their role in democratic backsliding. Taken together with Giglietto2025-ed60bc90’s overview of the post-DSA tool landscape, these papers suggest that the field’s central methodological problem — how to study platforms without their permission — has become inseparable from its central political problem: how to govern them at all.