A behavioural experiment with 500 participants, published in Nature Health on 1 May 2026 by a team at Julius Maximilian University of Würzburg, found that people described their symptoms in less detail when they believed an artificial-intelligence system, rather than a human doctor, would read the report. Descriptions intended for an AI chatbot averaged 228.7 characters, against 255.6 characters for those intended for a clinician, a difference of about 28 characters.

Participants wrote simulated symptom reports for two scenarios, unusual headaches and flu-like symptoms, under the belief that the reader was either a chatbot or a physician. The reduction persisted even among participants who were experiencing the relevant symptoms at the time. The researchers describe the effect as practically relevant: because AI triage and diagnostic tools depend on the detail a patient volunteers, systematically terser input raises the risk of an incomplete or incorrect assessment.

Two limitations temper the finding. The study measured simulated written reports rather than real consultations, and the absolute difference, roughly 28 characters, is small; whether it changes diagnostic accuracy in practice was not tested. The experiment also isolates a belief effect, what the patient thinks is reading the message, rather than any property of the AI itself. That is the variable digital-health services can most readily address, through disclosure and interface design that encourage patients to write as fully for a chatbot as they would for a clinician.