Google's AMIE AI Doctor Passes First Real Patient Test

An AMIE AI doctor just got its first real-world test with actual patients, and the early results are surprisingly solid. Google's diagnostic chatbot chatted with people before urgent primary care visits at a Boston hospital, and not one conversation had to be shut down for safety, according to a study published Oct. 8 in The Lancet.

Before you picture a robot replacing your GP, hold up. A human physician watched every single exchange live, ready to jump in. This was a feasibility test, not proof that AI improves health outcomes, and the researchers say so themselves.

What the AMIE AI doctor actually did

AMIE, short for Articulate Medical Intelligence Explorer, is a conversational system built by Google. Patients used it from home through a secure text chat after booking an urgent appointment with a physician at Beth Israel Deaconess Medical Center (BIDMC), according to Nanowerk's write-up of the BIDMC release.

The bot asked about symptoms, gathered medical history, suggested possible diagnoses for patients to discuss with their doctor, and produced a summary for the clinician to read before the visit. Think of it as a very thorough intake nurse that never gets tired.

The numbers behind the trial

The study ran between April and November 2025 and enrolled 114 patients. Of those, 98 finished both the AI chat and the in-person appointment. Every conversation was monitored in real time by a board-certified internal medicine physician.

Across all 98 completed encounters, none needed a safety-stop intervention. Supervising doctors flagged one hallucination and added clinical clarification in five cases. That is a small error footprint, though it is still not zero.

On accuracy, R&D World reports that AMIE's first seven diagnostic suggestions included the final diagnosis in 90% of cases. Its top pick matched 56% of the time.

Did doctors find it useful?

In 44 cases, physicians reviewed the AI transcript or summary before meeting the patient. In 75% of those, they said it helped them prepare, and in 57% it may have shaped how they handled the visit, according to BIDMC.

There was one cautionary tale. A provider called the interaction somewhat harmful in a single case because a patient may have felt anxious after AMIE listed lymphoma among possible diagnoses. Scary words from a chatbot land differently than from a doctor in the room.

Trust is the weak spot

Patients rated the chats favorably, praising how well the system listened, explained things and put them at ease. Their attitudes toward AI actually improved after the chat and stayed higher after seeing their doctor.

Still, two worries stuck around: whether their information stays confidential, and whether the chatbot is honest. Study author Adam Rodman of BIDMC said patient trust in medical AI tends to flow from relationships with their human providers.

Why this matters for your next doctor visit

Primary care is stretched thin. Aging populations, staffing gaps and increasingly complex patients all strain clinics, which is why researchers keep chasing tools that can take some of the load. Chatbots have aced simulations for years, but real patients are messier than test cases, and that gap is exactly what this trial tried to measure.

The authors published the work in The Lancet, one of the most closely read medical journals, under the title "Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study."

Co-senior author Marc Cohen of BIDMC said the future involves continued involvement of human clinicians, with a triad of patient, clinician and AI working together. That framing matters, because it treats the bot as a teammate that preps the appointment rather than a gatekeeper that decides your care.

The fine print you should know

This was one clinic and 98 completed patients, and Alphabet funded the research. Rodman was also a visiting researcher at Google during part of it, R&D World notes. Previous research has shown chatbots can stumble when everyday people drive the conversation, so this is a promising data point and not a verdict.

For Gen Z, the takeaway is simple: AI intake tools are heading to clinics near you, probably with a human still in the loop. If you want to follow more of this, browse our health coverage and our AI news for the next chapter.

The next step researchers describe is testing when human-on-the-loop workflows are appropriate, and whether the tech improves outcomes. Until then, the AMIE AI doctor is a clever sidekick for physicians, not a replacement for one.