Cullen, Z. N., Ellison, N. B., & Pasquetto, I. V. (2026). Seeing like an API: Platform-mediated research and the politics of access. Big Data & Society, 13. https://doi.org/10.1177/20539517261461463
Summary
This paper examines how academic researchers accessed, interpreted, and assessed data from Meta’s now-discontinued CrowdTangle (CT) API, drawing on interviews with 20 academic users plus supplementary interviews with former CT staff and power users. The authors introduce the “Seeing like an API” framework to argue that platform-controlled APIs function as epistemic infrastructures: their technical design, governance rules, and knowledge networks shape not only what data researchers can access, but what questions become legible, what methods are possible, and whose expertise is recognized. The central argument is that researchers exercise real but reactive agency—their workarounds simultaneously contest and entrench the platform’s epistemic terms. The paper lands its analysis in contemporary policy debates over third-party data governance, notably the EU’s Digital Services Act.
Key Contributions
- Introduces the “Seeing like an API” framework, synthesizing top-down (Scott’s Seeing Like a State) and distributed/emergent (Vertesi; Singh and Jackson) accounts of infrastructural visibility.
- Shifts analytic focus from merely diagnosing API limitations to documenting the situated methodological adaptations that data absences provoke, and their epistemic consequences.
- Provides a real-time empirical snapshot of researcher responses during CT’s deprecation—a record of a vanishing research infrastructure.
- Offers concrete governance recommendations: universal knowledge-sharing networks, transparent data-visibility decisions, support for verification/validation, and equitable, consistent access criteria.
- Argues for methodological pluralism and multiple access points enabling triangulation across APIs, data donations, and third-party intermediaries.
Methods
Semi-structured interviews with 20 academic researchers (each required at least one peer-reviewed CT-based publication) across eight countries, multiple disciplines, and career stages. Supplementary background interviews were conducted with three former CT team members and four non-academic early adopters/power users. Interviews were conducted remotely via Zoom between July and October 2021—before CT’s August 2024 shutdown—transcribed with Descript and analyzed in ATLAS.ti. The authors used a constructivist grounded theory approach with iterative open and focused coding, a data matrix to cross-check themes and surface negative cases, and sampling to saturation. The conceptual framework was developed through engagement with infrastructure studies literature.
Findings
- CT was frequently the only tool for large-scale access to organic content from public Facebook pages, groups, and Instagram profiles, yet participants described it as showing only “the tip of the iceberg.”
- Key limitations: exclusion of comment data, follower-count thresholds (criticized as U.S.-centric and excluding locally influential small accounts), restrictions on downloading media, and opaque content removal.
- Engagement metrics (likes, shares, reactions) stripped of discursive context could not disambiguate user intent, constraining interpretive claims.
- Three recurring adaptive strategies emerged: optimizing (community Slack channels, webinars, feature requests—a “common brain trust”), extending (web scraping, manual collection, integrating other APIs), and verifying (triangulating CT outputs against scraped/manual data, exposing bugs and discrepancies).
- Informal knowledge networks and API access were unevenly distributed, favoring elite institutions and those with proximity to the CT team; scraping workarounds were fragile, risked account bans, and were sometimes rejected by reviewers as terms-of-service violations.
- The rumored and actual shutdown was framed as destabilizing for research design, funding, teaching, and training, risking a fragmentation of the field toward bespoke, less replicable scraping.
Connections
This paper is a cornerstone of post-API / “APIcalypse” debates on platform-mediated research access, connecting closely to work on platform data access politics such as Freelon2018-ao, Rieder2025-ju, and Rogers2026-zz. Its concerns about scraping fragility, unequal access, and governance mechanisms like the DSA resonate with studies of alternative access infrastructures and data donation (e.g., Ohme2026-nv, Pierri2026-ib) and critical platform-governance scholarship including Gillespie2010-as, van-Dijck2018-up, and Petre2019-zh. The “Seeing like an API” framing also speaks directly to CrowdTangle-focused empirical and methodological work such as Bruns2019-nr.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify