Awareness, Disclosure, and Unverified Reliance on Generative AI for Health Information in Poland: A Nationwide Cross-Sectional Online Survey

Generative AI chatbots (e.g., ChatGPT, Gemini, Copilot, Claude) are increasingly used for health purposes. Unlike search engines, conversational systems select and synthesize information into a single personalized answer, and users may not recognize that part of the interpretive process has been transferred to the system. This study estimates the prevalence and co-occurrence of: limited awareness of how chatbot answers are produced, trust in non-diagnostic conversational cues, disclosure of health information, interpretive delegation, epistemic closure, and adherence without independent verification, among adults in Poland. The central question is whether users' level of reliance is calibrated to their understanding of the system on which they rely (miscalibrated epistemic trust).