pipette
ESEspañol

Active Health Problems in Real-World Consultations on a Purpose-Built Medical AI Platform

B. Adamson, A. W. Li, S. Gilson, B. Waxse, G. Ambwani, B. Crowe, A. Oskowitz

PreprintReal-world use

In the authors' words

Importance: Consumers use conversational artificial intelligence (AI) for health at scale, but whether purpose-built medical AI is used primarily for information seeking or as an entry point into care remains unclear. Objective: To characterize the clinical concepts users present, their visit intentions, and progression to physician telehealth care in real-world consultations on a purpose-built medical AI platform. Design: Exploratory retrospective cohort study of consultations sampled in equal monthly numbers between June 1, 2025, and June 30, 2026. Setting: A free, publicly available, consumer-facing, anonymous-by-default medical AI platform. Participants: Random sample of 26,000 English-language consultations by users aged 18 years or older, each with at least 5 chat interaction rounds. Exposure: A chat consultation with the medical AI. Main Outcomes and Measures: Clinical concepts linked approximately to SNOMED CT; visit intentions classified with a prespecified 11-category schema and reclassified with a published 12-category consumer-health taxonomy; completion of the AI consult with a structured SOAP note; and progression to a completed physician telehealth visit. Automated extraction pipelines were validated against physician annotation in a random subsample. Results: Of 26,000 sampled consultations, 56.7% were by users recorded as female and 50.9% by users aged 18 to 34 years. Among 25,975 consultations with successful extraction, 77,597 clinical concepts were identified; headache, abdominal pain, fatigue, and fever were the most common presenting concepts. Of 55,257 classified intentions, 60.1% (95% CI, 59.7%-60.5%) concerned a new or worsening condition. When consultations were reclassified with the published consumer-health taxonomy, 74.3% (95% CI, 73.8%-74.8%) included symptom questions or health concerns, whereas 5.0% (95% CI, 4.7%-5.2%) included health information or education, reversing the rank order published for a general-purpose assistant. Overall, 53.9% of sampled consultations ended with a structured SOAP note, and 15.1% of those progressed to a completed physician telehealth review. Conclusions and Relevance: Users of this purpose-built medical AI platform predominantly presented active personal health problems rather than general information requests. Symptom-driven use, generation of structured clinical documentation, and progression to physician care suggest that purpose-built medical AI may function as an entry point to clinical care for some users.

Main resultThe abstract does not state a limitation.

Appeared: Thursday, November 19. medRxiv. Preprint, not yet peer-reviewed.

DOI: 10.64898/2026.11.15.26368158