Conspiracy beliefs are widely considered resistant to factual correction, yet recent research shows that relatively brief, personalized “debunking” dialogues with a generative AI model can substantially reduce such beliefs. To identify the mechanisms driving this effect, we conducted an experimen…