AI could help inoculate us against misinformation

Embargoed until: Publicly released: 2026-08-05 09:01

International

Photo by Element5 Digital on Unsplash

AI has made it cheap and easy to spread misinformation online, but now US researchers say AI could also be a tool to help stop misinformation by 'inoculating' people against common myths. The researchers had AI write short "prebunking" articles for five common election myths, and then around 4000 US adults were shown one of these prebunking articles, then a persuasive piece pushing the myth. The researchers found that the AI-generated prebunking significantly reduced belief in election
rumours, with the effects still present one week later. They also helped offset declines in confidence in national election administration, and worked across party lines with no evidence of partisan backlash.

News release

From:

Toward Scalable AI-Assisted Prebunking of Election Misinformation: Evidence from a Preregistered U.S. Panel Experiment
Royal Society Open Science
Misinformation outpaces fact-checkers, and AI makes deception cheap. We show the same AI can inoculate against misinformation. In a preregistered study of 4,293 U.S. registered voters before the 2024 election, AI wrote short "prebunking" articles for five common election myths, drawing on a single human-refined prompt and trusted election-security sources. Each reader read one prebunking article, then a persuasive piece pushing the rumor. The intervention protected against changes in false-rumor belief and election-institution confidence, both immediately and a week later. It worked across party lines, with no backlash. Once the prompt was set, articles produced without further human review worked as well as ones with it. AI can spread misinformation; it can also protect against it.

Attachments

Note: Not all attachments are visible to the general public. Research URLs will go live after the embargo ends.

Research The Royal Society, Web page Please link to the article in online versions of your report (the URL will go live after the embargo ends).

Journal/
conference: Royal Society Open Science

Research:Paper

Organisation/s: California Institute of Technology, USA

Funder: R.M.A. and M.L.’s work on this project, and the data collection, was supported by a grant to the California Institute of Technology by the John Randolph Haynes and Dora Haynes Foundation.

Media Contact/s

Contact details are only visible to registered journalists.