Rogers, R., & Koronska, K. (2026). The performance of borderline content on Facebook. Content Moderation across Social Media Platforms, 97–123. https://doi.org/10.4324/9781003693192-4
Summary
This chapter traces how “borderline” content — material that approaches but does not cross Facebook’s community standards — has performed on the platform across the 2016, 2020, and 2024 U.S. presidential elections. Rogers and Koronska pair a historical critique of Meta’s progressively shrinking data-access regimes with an empirical replication of Craig Silverman’s 2016 BuzzFeed “fake news” study, updated for 2024 via off-platform tooling. They argue that the apparent decline of problematic content’s performance — with mainstream sources increasingly outperforming the fringe by 2024 — conceals a persistent pattern: the borderline content that does perform well remains almost exclusively right-wing. This outcome, they contend, is partly obscured by Meta’s rhetorical redefinitions of the problem (from “fake news” to “misinformation” to “borderline”), its shift from engagement to reach metrics, and its curtailment of independent research access.
Key Contributions
- A longitudinal, replicable comparison of problematic/borderline content performance across three U.S. election cycles (2016, 2020, 2024).
- A documented critique of Meta’s evolving moderation terminology and its drift toward “content-agnostic,” depoliticized moderation and “personalized ranking.”
- A “post-API”/post-CrowdTangle methodological pathway for studying Facebook using off-platform tools (BuzzSumo) and third-party labeling (MBFC, GATE).
- Empirical evidence that right-leaning borderline content persists despite overall decline, exposing tensions in Meta’s depoliticization narrative.
- An analysis of the politics of metric choice — reach vs. engagement — as an instrument that obscures the dominance of right-wing commentators.
Methods
The authors combine documentary and historical analysis of Facebook/Meta’s data-access regimes (Open Graph API, Graph Search, Social Science One, CrowdTangle, and the Meta Content Library) and moderation policy touchstones with an empirical replication of Silverman’s 2016 study for 2024. Using BuzzSumo as a data source, they built candidate- and issue-based keyword lists drawn from party platforms and New York Times headlines, ranked trending Facebook URLs by total engagement, and labeled domains via Media Bias/Fact Check (with custom borderline/biased categories). They searched over 12,000 unlabeled URLs for “pink slime”/imposter sources using Bengani’s list and the GATE service, and examined top-performing borderline YouTube channels as case studies.
Findings
- In 2024, borderline content drew only a small fraction of engagement (~330,723 of 47 million election-related engagements), far less traction than in 2016 or 2020.
- Mainstream domains (CNN, Fox News, BBC, NYT) consistently outperformed borderline domains; BBC and NYT dominated September–October 2024.
- Borderline content attracted attention only in short bursts driven by extreme-right sources (e.g., infowars.com, zerohedge.com), spikes that did not align with major campaign events.
- Engagement with borderline content came disproportionately from right-wing domains; biased content received significantly less engagement than less-biased content, a gap widening in the final six months.
- Center-left outlets received notably higher engagement than right-leaning counterparts just before the election.
- No imposter/“pink slime” sources were found among 12,000+ unlabeled URLs — echoing the 2020 result.
- The best-performing borderline content was largely on YouTube (~20 of the top 100 videos from borderline channels), and all supported Donald Trump.
- Meta’s ad revenue remained robust (over $39 billion in Q2 2024) despite efforts to reduce political content.
Connections
This chapter is an exemplar of “post-API” critical digital methods and sits closely with Rieder2025-ju and Rieder2026-pp on Meta’s steering and curtailment of independent research, and with Gonzalez-Bailon2024-rq and Bak-Coleman2025-pm on the methodological and political stakes of platform data access. Its empirical focus on how problematic content actually performs connects it to engagement- and exposure-based misinformation studies such as Allen2020-nj and Budak2024-ef, while its account of the definitional politics of “fake news”/“borderline” extends content-moderation scholarship including Gillespie2022-jx. It also pairs naturally with other work by the same author on platform governance, Rogers2026-cy.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify