AMIE reaches real patients—with physicians still supervising
Editorial draft · Reviewed October 15, 2026. Original source scope and limitations retained.
Prospective, single-centre, single-arm feasibility study
Google’s AMIE conversational AI has moved beyond simulated consultations into an early study involving real urgent primary care visits. The result is encouraging, but narrower than a claim that the system has been proved safe.
Of 114 enrolled adults, 98 completed both an AMIE interaction and their clinical appointment. Physicians monitored the conversations using predefined safety criteria. No conversation required a safety stop. Supervisors nevertheless recorded one hallucination and added clinical information in five interactions.
Clinicians completed post-visit surveys for 60 cases; in 44 they had reviewed the AI transcript beforehand. Of those 44, 33 found it helpful for preparation and 25 said it might have changed their behaviour. Those denominators matter: “75% helpful” does not describe every enrolled patient or every visit.
The study was published in The Lancet and funded by Alphabet. It had one centre and no comparison arm, so it cannot establish better outcomes or performance in unsupervised practice.
The near-term opportunity is structured history-taking that helps a clinician arrive better prepared. Larger comparative studies must test missed diagnoses, workload, patient outcomes and safety across different populations before routine deployment.