We recruited 2,193 U.S. adults online and conducted two experiments that systematically removed potential confounds. Personalized dialogues with an AI consistently reduced conspiracy beliefs. The debunking effect remained robust when we disclosed the AI’s persuasive goal, framed the exchange as a debate, instructed the AI to inform rather than persuade, halved the response length, altered belief measurement methods, or replaced the dialogue with a single tailored message. Only when the AI was instructed to persuade without factual evidence or arguments did the effect diminish substantially. Content analysis of the AI’s output and participants’ self‑reports linked larger belief reductions to factual argumentation, not to rapport building, storytelling, or empathy appeals. Thus, AI dialogues appear to lower conspiracy beliefs primarily by supplying relevant, well‑reasoned counter‑evidence. Our findings suggest that humans could achieve comparable debunking outcomes if they can provide evidence of similar quality.
Review: The study highlights the pivotal role of evidence‑based reasoning in correcting false beliefs and offers empirical guidance for designing more effective science communication and misinformation‑countering tools.