LLMs Successfully Increase or Decrease Conspiracy Beliefs via Prompting
July 17, 2026
Experiments with nearly 4,000 Americans show frontier models can effectively promote or debunk conspiracy theories. Users in the bunking condition reported higher trust and perceived the AI as more collaborative than those in the debunking condition.
HOW THIS AFFECTS YOU
●
builderYou should be aware that model helpfulness/trust metrics may correlate with the ability to spread misinformation.
●
policyYou must consider how model persuasion capabilities impact social stability and misinformation.