Evaluation of Large Language Models in Traditional Chinese Medicine: Clinical Reliability and Lay Perception of DeepSeek Diagnoses in Cardiovascular Disease

Large language models (LLMs) are increasingly used for health information, yet little is known about how AI-generated advice is received within traditional medical systems. This study protocol examines Traditional Chinese Medicine (TCM) in mainland China through two linked phases. In the first, we will submit the self-reported symptoms of 20 cardiovascular outpatients from a tertiary TCM hospital to DeepSeek-V3 and have three independent, blinded TCM practitioners rate the model’s diagnoses and herbal prescriptions against the attending physicians’ records. In the second, an online survey of the general public will present paired, anonymized diagnostic texts, one written by a physician and one generated by DeepSeek. Participants will judge which text came from a human physician and rate each on professionalism, trustworthiness, warmth and willingness to follow the advice. The study asks what happens to the social foundations of therapeutic trust when a healing tradition grounded in the embodied authority of the practitioner can be convincingly voiced by a machine.