Cazzamatta, R. (2026). From moderation to chaos: Meta’s fact-checking and the battle over truth and free speech. New Media & Society. https://doi.org/10.1177/14614448251413687
Summary
This paper interrogates the empirical basis of Mark Zuckerberg’s January 2025 decision to end Meta’s Third-Party Fact-Checking Program and his characterization of fact-checkers as agents of censorship. Combining a comparative content analysis of 2,053 debunking articles from 16 Meta-partner organizations across eight European and Latin American countries with 30 expert interviews, Cazzamatta demonstrates that the censorship narrative is empirically unfounded: fact-checkers never had authority to remove or downrank content, and most verified false content stays online, either labeled or entirely unaltered. The study argues that Meta’s shift from “preemptive” to “conflictual” cooperation signals a turn toward politically aligned, oligarchic moderation that threatens evidence-based regulation—especially in under-regulated regions like Latin America.
Key Contributions
- First empirical evidence on what actually happens to content after fact-checker verification on Meta platforms, directly testing censorship claims.
- Cross-regional comparison spanning the two regions (Europe and Latin America) named in Zuckerberg’s speech.
- Triangulation of large-scale content analysis with practitioner interviews.
- Documentation of fact-checkers’ normative positions on free speech, transparency, and moderation.
- Extension of Rone’s framework of preemptive versus conflictual platform cooperation to the dismantling of the fact-checking program.
- Empirically grounded skepticism about replacing professional fact-checking with Community Notes in polarized, under-regulated contexts.
Methods
- Comparative quantitative content analysis of 2,053 debunking articles (2022) from 16 Meta-partner organizations across eight countries (Germany, Portugal, Spain, UK, Argentina, Chile, Brazil, Venezuela), stratified from 11,636 collected links.
- Inductively developed codebook (via a 500-article pre-test) capturing post-verification outcomes: warning labels, removal, court orders, no action, residual footprint, source pressure, voluntary corrections.
- Coding by eight native-language assistants over eight months; intercoder agreement 77%–93%.
- Correspondence analysis mapping country–outcome associations (85% of variance).
- 30 semi-structured expert interviews (Aug 2024–Mar 2025), analyzed inductively in NVivo following Tracy’s qualitative textual analysis.
- Country selection grounded in Hallin and Mancini’s media system typology and levels of deliberative democracy plus market size.
Findings
- Correspondence analysis found no strong country-specific patterns—most outcomes clustered near the null point, indicating cross-national uniformity.
- Content removal ranged from 21% (Venezuela) to 45% (Portugal), clustering around 30%, but attribution to Meta versus the original spreader is often unclear.
- Platform labeling varied widely, from 10% (Portugal) and 0% (Argentina) to 56% (Brazil).
- In many cases no action was taken and false posts remained accessible without labels, undermining censorship narratives.
- Successful correction requests from public figures were rare across all countries.
- Fact-checkers marshal ten key arguments against removal, foremost protecting free speech while distinguishing it from “freedom of reach,” and worry that opaque takedown criteria fuel conspiracy theories.
- European fact-checkers welcome the Digital Services Act and EU-level advocacy (EFCSN); Latin American fact-checkers are more pessimistic about regulation and funding.
- Fact-checkers fear Community Notes can be weaponized via coordinated voting, citing experiences on X.
Connections
This paper sits within work on the political economy and governance of fact-checking, resonating with studies of Community Notes and crowdsourced verification such as Allen2025-ot and DeVerna2025-dl, as well as analyses of platform moderation and data-access regimes like Rieder2025-ju and Rieder2026-pp. Its concern with counter-speech, labeling, and misinformation correction connects to the debunking and inoculation literature represented by van-der-Linden2026-jt and Lewandowsky2026-ob.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts