Confronting Misinformation Produced with Generative AI: A Meta-Analysis of Experimental Scientific Evidence

The IPIE's 2026 synthesis report is the most rigorous available meta-analysis of experimental evidence on how generative AI-produced misinformation affects audiences and what interventions work against it. Drawing on 60 effect sizes from 24 publications covering 33,801 participants, the report documents a divergence in public responses across modalities: perceptions of textual GenAI misinformation have become more credulous over time, while perceptions of visual GenAI misinformation (primarily deepfakes) have become more sceptical. The pattern is associated with the timing of data collection rather than underlying model capacity — suggesting that public calibration to AI content is shifting as exposure accumulates.

The countermeasures analysis is the report's most actionable contribution. Corrective information yields small-to-moderate reductions in perceived accuracy of GenAI misinformation in studies conducted after 2020, with pooled estimates that are significant and comparatively stable. The finding that content labelling shows inconsistent effects across contexts is important for platform governance debates: it complicates the assumption that labelling alone constitutes an adequate response to AI-generated misinformation. For researchers, policymakers, and practitioners working on AI content governance, this synthesis report represents the current scientific consensus on what experimental evidence actually shows.