Analysis of 1.76M Community Notes received by X from January 2021 to March 2025: 90%+ of submitted notes are never published, with many notes “stuck in limbo”
Context & Ripple Effects
This analysis extends a persistent coverage gap: a 2022 review found most contributed notes were invisible, while a later election-claims study found few responded-to posts ultimately displayed a note. X’s scoring-system redesign promised faster publication, but the new dataset suggests speed improvements have not resolved the larger question of which submissions ever clear the threshold.
The finding arrives as participation has weakened, with monthly Community Notes submissions falling sharply earlier in 2025. It also gives added weight to X’s plan for AI agents to propose notes while people rate them: generating more candidate notes will not by itself address a publication pipeline in which many remain unresolved.
First-order effects
- Readers see only a small fraction of the corrective context contributors submit, limiting the feature’s practical reach on disputed posts.
- Contributors face an opaque outcome for many submissions: rather than being clearly published or rejected, notes can remain pending without reaching the audience they were intended to inform.
Second-order effects
- X’s prior scoring redesign aimed at getting notes onto posts faster is judged not just by time-to-display, but by whether it can resolve a far larger pool of candidate notes.
- AI-written note proposals could increase supply, but would intensify pressure on the human-rating and consensus process if unresolved submissions are already the binding constraint.
Third-order effects
- The pattern underscores a trade-off in consensus-based moderation: it can seek broad agreement before attaching context, but that standard can constrain coverage and timeliness when participation or consensus is limited.
- If the backlog persists, Community Notes may evolve toward systems that prioritize ranking and resolving candidate context—not merely producing it—while retaining human judgment as the credibility safeguard.
The trend: Community-led fact-checking is shifting from a problem of generating annotations to one of efficiently reaching enough cross-user agreement to publish them.