Proceedings of the Annual Meeting of the Cognitive Science Society

Vaccine conspiracy beliefs contribute to vaccine hesitancy and resistance to public health interventions. Recent work suggests that large language models (LLMs) may reduce conspiracy beliefs through personalized, evidence-based dialogue. The present study extends this work by examining whether LLM-mediated conversations reduce vaccine conspiracy beliefs and whether effectiveness depends on conversational style in a country characterized by negative attitudes towards vaccination. Participants engaged in an interactive conversation with an LLM following open-ended elicitation of their vaccine conspiracy beliefs. The LLM used rational, emotional, or free-argumentation, or a neutral control conversation. Belief in vaccine conspiracies was assessed before and after the interaction using summaries of the participant's personally endorsed vaccine conspiracy beliefs. Results showed a reduction in vaccine conspiracy beliefs only when rational argumentation is used. However, no effect was observed for vaccine-related behavioral intentions. These findings confirm the conclusions of previous studies that belief change via AI requires rational engagement.