#1292 2021 · Twitter (Birdwatch, later Community Notes) · Social media / content moderation
Twitter stopped picking who fact-checks tweets; required disagreeing raters to agree first
the problem
Company fact-checkers couldn't review tweets at scale, and any reviewer's call was read by someone as proof of bias
background
Platform-run fact-checking faced two compounding problems: the sheer volume of tweets made comprehensive review by any employed team impossible, and whenever a company-appointed reviewer did label or remove a tweet, the decision itself became a flashpoint, read by one side or the other as proof of institutional bias. Simply opening the labeling process to crowd voting wouldn't fix the second problem, since a majority-rules vote on a contested claim tends to just reproduce whichever political group has more users voting, reinforcing rather than resolving the perception of a biased outcome.
Twitter's Birdwatch team, launched in 2021, built a different selection rule into the system itself. Any user could propose a correcting note on a tweet, but a note would only surface publicly if it was rated helpful by contributors who, based on their voting histories, don't normally agree with each other — a 'bridging' algorithm that treats cross-ideological consensus, not raw vote count, as the bar for visibility.
what everyone would do
The available options were to keep the review inside a company-employed trust and safety team, whose calls got read as institutionally biased regardless of outcome, or open corrections to plain crowd voting, which just lets whichever political faction has more active raters set the record.
what they saw
The team saw the credibility problem was not who did the rating, but what kind of agreement counted. A majority-liked note proves little; one that normally-opposed raters both call helpful is hard to dismiss.
the move
Birdwatch, renamed Community Notes, let any qualifying user write a proposed correction to a misleading tweet, then used a bridging algorithm to promote only the notes rated helpful by raters who have historically disagreed with one another, rather than by simple majority vote — making cross-ideological agreement itself the filter for what the public sees.
why it works
By scoring notes on whether raters with historically divergent voting patterns both marked them helpful, the algorithm filters out corrections that only appeal within one ideological cluster, even if that cluster is large enough to win a simple majority vote. This shifts the selection pressure from popularity to cross-cutting credibility, which is why notes that clear the bar measurably move viewer belief and sharing behavior across the political spectrum rather than only within it — the mechanism is doing real epistemic work, not just distributing the moderation workload.
the payoff
Notes visible to all US users by 2022 made viewers 20-40% less likely to believe a misleading tweet and 15-35% less likely to share it.
where it breaks
It requires enough raters from genuinely different perspectives to be active on the same content for a bridging signal to exist at all; on low-traffic posts, or in communities without real ideological diversity among participants, there's no cross-group disagreement to bridge, and the system either produces no note or falls back toward ordinary majority dynamics.
what came after
Community Notes became the most widely studied large-scale deployment of a bridging-based ranking algorithm for content moderation, cited in academic research on reducing polarization in crowd-sourced fact-checking and adopted as a model discussed by other platforms building their own community moderation tools.
references
- [1]Twitter is making its crowdsourced fact-checks visible to all U.S. users with Birdwatch expansionTechCrunch, 2022techcrunch.com
- [2]Helpful Birdwatch notes are now visible to everyone on Twitter in the USX (Twitter) official blog, 2022blog.x.com