Correcting Health Misinformation Using AI Chatbots: Effects of Human–AI Collaboration Mode, Transparency, and Uncertainty

This project examines how the presentation of human–AI collaboration, scientific uncertainty, and supporting explanations shapes public acceptance of AI chatbots for health misinformation correction. Guided by the Human–AI Interaction framework grounded in the Theory of Interactive Media Effects (HAII–TIME), the study investigates whether these features independently and jointly influence trust in a chatbot, attitudes toward using it to correct health misinformation, and intentions to use it for this purpose. The study employed a 3 × 2 × 2 between-subjects online experiment with a final analytical sample of 424 participants in China. Participants were randomly assigned to one of 12 conditions and viewed a simulated chatbot conversation addressing the false claim that insulin injections are addictive. Human–AI collaboration was described as human-dominated, with experts leading the evaluation and AI assisting; AI-dominated, with AI leading and experts verifying; or AI-only, without human involvement. Correction uncertainty varied according to whether the accompanying notice described the scientific evidence as broadly consistent or unresolved. Correction transparency varied according to whether the chatbot disclosed supporting reasons and sources. Analyses examined the main and interaction effects of these features while accounting for demographic characteristics, AI literacy, eHealth literacy, and perceived risks of AI hallucinations.