The Google AMIE primary-care study, published in The Lancet on October 8, 2026, tested supervised conversations before clinic visits. Patients enrolled from April through November 2025 at Beth Israel Deaconess Medical Center (BIDMC); 98 of the 114 enrolled completed both an AMIE conversation and an appointment with a primary-care provider (PCP).
A real-clinic study of supervised AMIE
Google and BIDMC conducted a prospective, single-center, single-arm feasibility study. In plain English, researchers followed a planned study in one clinic to see whether the approach could be carried out; the single-arm design had no comparison group.
AMIE—short for Articulate Medical Intelligence Explorer—is a conversational diagnostic AI. The study examined its use as preparation for a visit, not as a stand-alone clinician.
How patients and clinicians used AMIE
Participants were English-speaking adults with one complaint and an urgent primary-care appointment. They could talk with AMIE up to five days before the visit. Physician safety supervisors monitored every interaction and were trained to intervene under predefined criteria. Before the appointment, each PCP received the chat transcript and a summary.
Completion and safety observations
Of 114 patients enrolled, 98 completed both the AI interaction and the PCP appointment. No conversation met the prespecified criteria for a safety stop. Supervisors noted one hallucination and added clinical information in five interactions.
Clinical evaluators rated AMIE’s conversations favorably in 87–100% of cases across 17 criteria. Patients’ favorable ratings ranged from 48–96% across 16 criteria. These ranges reflect ratings across different criteria, rather than one overall score. Patient attitudes toward AI improved after the AMIE conversation and remained elevated after the physician visit.
What PCPs said about AMIE’s summaries
PCPs completed post-visit surveys in 60 of the 98 completed cases. In 44 cases, they had reviewed the AMIE transcript before the appointment. In that 44-case group, PCPs said the summary helped them prepare in 33 cases; in 25, they said AMIE might have changed their behavior. The second result describes a possible change, not a definite one.
What a feasibility study can establish
The study evaluated supervised, pre-visit support in one clinic, with a PCP appointment still part of each completed patient’s experience. It did not test autonomous diagnosis or treatment, or whether AMIE could replace primary-care clinicians. The study calls for further research.
