Abstract
<title>Abstract</title> <p>The public increasingly uses generative AI to obtain health advice, yet it remains unclear whether preferences for AI versus physicians mainly reflect attitudes toward AI or vary across health-advice use cases. This preregistered between-subjects experiment examined preferences for AI versus physician-based advice as first and second opinions across nine health system utilization use cases in a UK sample representative for age, gender, and region (N = 1,799). Participants were randomly assigned to one of 27 real cases sampled from online health questions and nested within the nine use cases. Overall, most participants preferred physicians for a first opinion (83.2%), but AI preference differed substantially by use case, ranging from 4.0% for acute symptom management and care-seeking decisions to 28.5% for post-incident infection risk. For second opinions, physician-based advice double-checked by another physician was most common (49.0%), followed by AI advice double-checked by a physician (34.4%), whereas receiving both opinions from AI was rarely preferred (0.9%). Use-case differences remained significant after adjusting for general attitudes toward AI. These findings suggest that preferences for AI-generated health advice are not reducible to general pro- or anti-AI attitudes but vary with the type of health question and whether AI is used alone or within a physician-validated advice pathway.</p>