Evidence may work better than labels against misinformation

Misinformation has become harder to counter as social platforms reward emotionally provocative content and generative AI makes credible fake articles, synthetic images and deepfake video cheap to produce. Warning labels and disputed-content flags have shown mixed results, in part because direct corrections can trigger psychological reactance and make people defend existing beliefs more strongly.

Abayomi Baiyere of Smith School of Business and collaborators at Copenhagen Business School tested a different approach called discursive evidence, which gives people material to evaluate a claim rather than telling them whether it is true or false. In experiments involving more than 1,000 participants, people assessed 22 news claims with no help or with curated evidence from PolitiFact. Clear, relevant evidence improved accuracy by around six to seven percentage points on average, including on politically charged claims that challenged prior beliefs.

The approach appeared strongest when participants were first primed to think critically. The researchers suggest a scalable system could use search tools to gather evidence, large language models to filter for relevance, and brief critical-thinking prompts embedded in platform interfaces. Other AI tools, including deepfake detection, bot-network analysis and content provenance systems, may help, but platform economics still favor the spread of misinformation over its disruption.