What they did
Three MIT researchers tracked roughly 126,000 story cascades — chains of retweets from a single origin — spread by about 3 million people more than 4.5 million times on Twitter, from 2006 to 2017. Every story was classified as true or false using six independent fact-checking organizations, which agreed with each other 95–98% of the time.
What they found
- False stories were 70% more likely to be retweeted than true ones, even controlling for account age, activity level, and follower counts.
- Truth took roughly six times as long as falsehood to reach 1,500 people.
- The top 1% of false-news cascades reached between 1,000 and 100,000 people; true stories rarely broke 1,000.
- False political news spread fastest, deepest, and most widely of any category — faster than false news about terrorism, natural disasters, science, or finance.
- The researchers identified and removed bot accounts from the analysis. The gap barely changed — bots amplified true and false stories at roughly equal rates. Humans were the ones spreading falsehood faster.
"Falsehood diffused significantly farther, faster, deeper, and more broadly than the truth in all categories of information... The degree of novelty and the emotional reactions of recipients may be responsible for the differences observed." — Vosoughi, Roy & Aral, Science 2018, abstract
The researchers' own explanation
The paper's leading hypothesis is that false stories are simply more novel — they say something surprising, which is precisely what a network built to reward engagement will amplify — and that novelty tends to provoke faster emotional reactions like surprise and disgust, which drive faster resharing than the calmer reactions typical of confirmed, unsurprising true stories.
Worth knowing
This is a Twitter (now X) dataset covering 2006–2017. Platform design, moderation policy, and user behavior have all changed since — on that platform and others — so treat the specific percentages as a snapshot of that decade rather than a permanent constant. The core mechanism the paper proposes (novelty and emotional reaction drive faster sharing) has since been examined on other platforms with generally consistent results, though exact figures vary by study and platform.