Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study

Key Findings

• — 114 patients were enrolled from April to November 2025 and 98 completed both the AMIE conversation and the physician appointment.
• — Zero conversation safety stops were required under the prespecified criteria; supervisors noted one hallucination and added clinical information in five interactions.
• — Clinical evaluators rated conversations favourably in 87-100% of cases across 17 criteria; patients rated them favourably in 48-96% across 16 criteria, and patient attitudes towards AI improved after the conversation.
• — Primary care physicians reviewed the AMIE transcript before 44 visits, found it helpful for preparation in 33 and said it might have changed their behaviour in 25.

Methodology Notes

Prospective, single-centre, single-arm feasibility study (ClinicalTrials.gov NCT06911398); English-speaking adults booked for a single-complaint urgent primary care visit interacted with AMIE by text up to five days before the appointment; all conversations supervised live by a physician. No control arm; not powered for diagnostic or outcome comparisons; funded by Alphabet, with Google-affiliated authors. Enrolment ran April to November 2025 (days not stated; observation dates set to the first and last day of those months). Published online 2026-10-08 (PubMed article date; PMID 42849491); 48 authors (first, last and Google leads listed). Verification route: thelancet.com is behind a bot wall; the structured abstract was read from PubMed and the Crossref record. A preprint version (arXiv 2603.08448) appeared in March 2026. An accompanying Lancet Comment (10.1016/S0140-6736(26)01763-0) was published the same day.

Sources