Study: less than 10% of users who retweeted false info during disasters removed their RT later, less than 20% of them clarified false RT with a new tweet
Study examines Boston Marathon bombing, Hurricane Sandy; also finds most users fail to correct misinformation
Context & Ripple Effects
This study closes the loop opened by earlier disaster-Twitter research: an analysis of 10M tweets showed they could predict Hurricane Sandy damage better than FEMA models, establishing Twitter as a genuine crisis-information channel. The same year, a separate study of 4.5M+ tweets confirmed inaccurate news spreads faster and further than true stories, with humans rather than bots doing the spreading.
What was missing was the back end of that lifecycle — what happens after a user amplifies something false. By examining retweet behavior around the Boston Marathon bombing and Hurricane Sandy, this study finds the correction rate is negligible: fewer than 10% delete the false retweet, fewer than 20% post a clarification. The false signal stays live in followers' feeds indefinitely.
First-order effects
- Followers of the roughly 90% of users who never remove a false disaster retweet keep seeing the misinformation with no marker that it was wrong — during exactly the events where accurate information matters most.
- The small minority who do correct mostly choose clarification over deletion, meaning even 'corrected' threads retain the original false claim at their core.
Second-order effects
- Platforms cannot rely on user self-correction, so the burden falls on top-down labeling — yet related coverage shows that mechanism underperforms too: fact-check labels on Trump tweets did not stop them from spreading further than unlabeled ones, and years later 59% of rated-false COVID content still sat on Twitter with no warning label at all.
- Emergency agencies and newsrooms lose the assumption that posting a correction reaches the people who saw the original error, forcing duplication of effort across channels rather than in-feed repair.
Third-order effects
- Combined with evidence that 0.1% of users shared nearly 80% of the fake news consumed in 2016, the pattern points at a structural fix: interventions aimed at the small population of repeat sharers — friction, rate limits, account-level consequences — rather than per-post moderation of content nobody retracts.
- If uncorrected amplification is the norm, regulators and platform designers will increasingly treat the retweet itself as the unit of harm, shifting liability debates from publishers to amplifiers.
The trend: Misinformation research has moved from proving falsehoods spread faster to documenting that almost nothing corrects them afterward, pushing platforms toward sharer-side friction instead of trusting user retractions.