Auditing the impact of social media's policy shift on anti-vaccine discourse: A large language model-driven empirical study
Source check: Source checked. Not independent scientific confirmation.
Finding: GPT-4o-assisted annotation of 13,458 COVID-19 vaccine-related tweets enabled a short-window analysis of discourse around X ending its misinformation policy. Anti-vaccine tweets increased after the policy change (OR 1.60, 95% CI 1.50–1.72), largely through retweets; health-concern themes rose while conspiracy and anti-mandate shares declined. The 14-day observational pre/post comparison does not establish causation or long-term effects.
Technology: GPT-4o classified tweet stance, assigned anti-vaccine themes, and assessed stance consistency; the validated labels enabled analysis of changes in discourse patterns.
Family GPT-4: profile · models page
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- GPT-4o classified tweet stance, assigned anti-vaccine themes, and assessed stance consistency; the validated labels enabled analysis of changes in discourse patterns.
- Test and result
- The paper reported external CAVES benchmark F1 of 0.883 for anti-vaccine stance and agreement checks, then analyzed a 14-day pre/post tweet sample.
- Limitation / still unknown
- Short observational pre/post window cannot isolate the platform policy as the cause or support conclusions about long-term discourse.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0346568
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 22 Nov 2026. Review scope: Primary full text, including Methods and Results; publication date and first result-bearing version checked..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.