JMIR Publications

ABSTRACT

Personal health large language models (PH-LLMs), exemplified by ChatGPT Health and Fitbit’s Gemini-powered health coach, are emerging consumer-facing systems that synthesize multimodal patient-generated health data into action-oriented narratives, reshaping how patients enter clinical encounters. This shift transforms the traditional dyadic clinician–patient relationship toward an emerging triadic model of negotiated authority in which clinicians must mediate among clinical evidence, patient values, and algorithmic narratives. This paper analyzes how PH-LLMs redistribute epistemic authority and proposes a clinical governance framework to preserve safety, accountability, and trust. Analyzing the intersection of emerging artificial intelligence capabilities and clinical workflows, we identify key opportunities: improved patient participation through structured sensemaking, between-visit support enhancing care continuity, and more collaborative encounters where clinicians function as accountable adjudicators. We also characterize three failure modes: authority conflict when recommendations diverge from clinical judgment, fragmentation of clinical truth across incompatible platforms, and diffusion of accountability across stakeholders. To address these risks, we propose a three-layer governance framework spanning evidence and provenance, clinical arbitration and workflow integration, and competence and accountability. When appropriately governed, PH-LLMs may become enabling infrastructure for safer, person-centered longitudinal care.