Physician Responses to Patient Use of Generative AI for Self-Diagnosis: A Field Experiment

2025-11-20

2026-01-20

Physician decision quality and decision-making processes during online consultations

Decision quality is measured by the extent to which physicians’ diagnostic assessments and recommended actions align with the ground-truth diagnosis and the clinically desirable care plan (e.g., diagnosis, recommended treatment, or the need for escalation to in-person care). Decision-making processes are measured using behavioral and textual indicators, including response time, hesitation or delay in acceptance, and the informational content and structure of physicians’ written responses.

Physicians are randomly assigned to one of three conditions that differ only in the content of the initial inquiry. After the initial inquiry, GenAI-based standardized patients conduct follow-up conversations with physicians. All patient profiles are otherwise held constant, and physicians were randomly assigned to one of three conditions: a control group, an AI-diagnosis condition, and a friend of a friend's recommendation condition.

Not available

Randomization conducted in the office by a computer.

Physician

No

Between 3 and 10 multiples of 96 physicians (i.e., 288–960 physicians).

Between 3 and 10 multiples of 96 physicians (i.e., 288–960 physicians).

Between 3 and 10 multiples of 32 physicians for each of the three groups: the control group, the AI-diagnosis disclosure group, and the human-source diagnosis disclosure group.