Before you see a doctor, what if an AI could take your history, ask about your symptoms and hand your doctor a neat summary? Google has tested exactly that with real patients. A study led by researchers at Google and Beth Israel Deaconess Medical Center (BIDMC) in Boston, published in The Lancet on October 8, 2026, looked at how Google’s research chatbot AMIE performed in a real primary care clinic.
Google says it’s the company’s first paper in the main Lancet journal. The results are encouraging, but this was a small, early study, and the details matter.
How the Google AMIE study worked
AMIE stands for Articulate Medical Intelligence Explorer, a research chatbot built by Google. Between April and November 2025, adults booked for urgent, non-emergency visits at BIDMC’s primary care practice were invited to chat with AMIE by text before their appointment. According to Becker’s Hospital Review, 114 patients enrolled and 98 completed both the AI chat and their appointment.
Safety was built in. A physician watched every conversation live and could stop it under predefined conditions, such as risk of harm or patient distress. Google says not a single conversation needed to be stopped. Doctors then received AMIE’s summary before the visit.
What the results show
The final diagnosis was decided by chart review weeks after each visit. According to the study, as reported by Unbiased Headlines:
- The final diagnosis was in AMIE’s list of possible diagnoses 90% of the time (88 of 98 cases)
- It was in AMIE’s top three in 75% of cases
- It was AMIE’s first choice in 56% of cases
Google says clinicians reported that the AI summaries helped them prepare for visits in 75% of cases and influenced their approach to care in more than half. The trial is registered as NCT06911398, and Unbiased Headlines notes that a preprint of the results first appeared in March, so the Lancet paper adds peer review to findings that had been public for months.
Two sides of the story
On one hand, this is a meaningful step. Most earlier AI medical chatbot tests used simulated patients or actors. Here, real people with real problems used the tool, with no safety stops. Google says AI like this could help ease the strain on healthcare workers and improve the patient-doctor relationship.
On the other hand, the limits are important. It was a single-site, single-arm feasibility study with no comparison group, and every chat had a doctor watching. It did not measure whether patients ended up healthier. The headline 90% figure means the right answer was somewhere on a list of up to seven options, while AMIE’s top guess was right about half the time. Not every doctor read the AMIE summary either: only about 61% of primary care doctors completed their survey. Google itself says larger trials are needed. Patient privacy, over-reliance on AI suggestions and who is responsible if an AI misses something serious are questions that will need clear answers before any wide rollout.
What this could mean for patients
As a writer, I feel this kind of AI could be most useful as a helper, not a replacement. In India, many doctors see dozens of patients a day with only a few minutes each, so a good pre-visit summary could save precious time. But I’d want patients to know clearly when they’re talking to an AI, and I’d want doctors to always make the final call. The careful, supervised setup in this study is the right way to start, and I hope future trials keep that level of safety.
For more health and science stories, read how market research turned lab science into a nasal spray, why reading for pleasure is linked to better health and how menopause is tied to brain changes.