The message arrived on a Tuesday morning: a voice note, allegedly from a doctor, warning that a common painkiller had been recalled. By evening it had reached tens of thousands of phones across four countries. It was entirely false.
Following the trail
Using a dataset of public messaging groups, my team reconstructed the claim's path hour by hour. It didn't spread evenly. It jumped between communities through a small number of “bridge” members — people who belonged to several groups at once, often trusted community figures.
These bridges weren't malicious. Most forwarded the message out of genuine concern. That's what made it so effective: it traveled on the credibility of people who meant well.
Misinformation rarely looks like a lie. It looks like a friend trying to help.
Why fact-checks arrive late
The first published fact-check appeared 31 hours after the original message. By then, nearly 80% of the eventual spread had already happened. Fact-checkers weren't slow; they were checking. The rumor didn't have to.
What seems to work
- Short, forwardable corrections in the same format as the rumor — often a voice note.
- Working with the bridge members themselves, who can reach communities fact-checkers can't.
- Pre-bunking: warning people about common tactics before a specific rumor arrives.
No single intervention stops a rumor. But understanding its anatomy tells us where to apply pressure — and who we need on our side.