Generative AI is reshaping the information landscape but the evidence on its effects and the effectiveness of countermeasures has, until now, remained fragmented.
The IPIE’s latest synthesis report, Confronting Misinformation Produced by Generative AI: A Meta-Analysis of Experimental Scientific Evidence, addresses this gap. Drawing on 60 effects from 24 publications and 33,801 participants, the report synthesizes experimental evidence on how AI-generated misinformation affects individuals’ ability to evaluate information and what interventions can mitigate its influence.
Key findings:
- Textual misinformation poses growing persuasive risks. More recent studies report higher perceived accuracy and credibility of AI-generated text, while skepticism toward visual misinformation, including deepfakes, has increased over time.
- Preventative corrective information is the most consistently effective countermeasure. Post-2020 studies show small-to-moderate reductions in perceived accuracy when corrections are delivered prior to exposure.
- Content labeling yields highly variable outcomes. While labels are associated with modest average reductions in perceived credibility, effect sizes vary widely across modalities, design, and contexts, and should not be treated as reliable standalone interventions.
These findings carry significant implications for platform governance, regulatory design, and public information policy.
A recorded expert panel discussion, held on 26 May, 2026, is also available for those wishing to explore the report’s findings and their policy implications in greater depth. Watch the recording.
News Item