Patient-Centered Communication Preferences in AI-Powered Mental Health Chatbots: Evidence from Two Preregistered Studies

Access to mental health information is shifting from static search to conversational AI. Guided by patient-centered communication (PCC), two preregistered U.S. studies identified preferred communication features for interactions with AI chatbots about mental health and how individuals trade them off within feature bundles. Study 1 (N = 414, US quota sample) used a Best–Worst Scaling (BWS) to identify the six most relevant PCC-aligned features for healthcare providers. Study 2 analyzed an AI chatbot subsample (n = 268) drawn from a U.S. quota-representative sample in a Discrete Choice Experiment (DCE) to quantify trade-offs between combinations of these preferred features. Across both studies, users strongly wanted two communication features simultaneously in AI mental-health chatbots: reflective listening and multi-symptom assessment. Importantly, relational and clinical PCC-aligned features are most highly valued in interactions with AI mental-health chatbots. These preferences remained largely consistent across users and their preferences for communication accommodation.