Updated October 12, 2026. A Google and BIDMC clinical study of AMIE involved 98 patients. Here’s what the October 2026 Lancet report found, and its limits.
A real-world look at AI before medical visits
On October 8, 2026, Google discussed research conducted with Beth Israel Deaconess Medical Center and published in The Lancet. In the clinical study, 98 patients spoke with AMIE, a research diagnostic AI chatbot, before urgent primary-care visits. Supervising physicians monitored the conversations in real time. The study is interesting because it investigates a concrete clinical workflow: whether an AI-led pre-visit discussion can help a doctor arrive better prepared.
What the reported results actually show
According to Google’s summary, none of the 98 monitored conversations required intervention under the study’s predefined safety criteria. Clinicians said the AI-generated summaries helped them prepare in 75% of cases and influenced their care approach in more than half. Google also reported that AMIE’s differential diagnoses matched doctors’ final diagnoses 90% of the time. These are findings from a relatively small, supervised study, not proof that an autonomous chatbot is ready to diagnose patients independently.
Why the pre-visit workflow matters
A busy medical consultation often begins with collecting symptoms, medications, history and the patient’s immediate concerns. If a supervised AI system can organize that information clearly, clinicians may spend more time on examination, interpretation and shared decision-making. Better preparation could also help patients remember questions they otherwise forget. However, summarization errors or missing details could mislead a clinician, so a doctor must remain able to check and correct the record.
Safety and evidence limitations
The sample size was 98, and the study occurred within a particular care setting. Results may not transfer directly to emergencies, complex chronic illness, different languages or clinics with fewer resources. It is important to distinguish diagnostic agreement from improved health outcomes: matching a final diagnosis does not by itself establish better recovery, lower cost or fewer adverse events. Larger studies should examine representative populations, bias, privacy, patient trust and long-term outcomes.
What health organizations can learn
Hospitals considering AI intake tools should prioritize informed consent, protected health data, clear escalation routes and continuous quality audits. Staff need visibility into source information instead of relying on generated summaries alone. Patients should know that a research system does not replace examination or qualified medical care. The most promising near-term applications may be those that strengthen the clinician-patient relationship without weakening accountability.
Frequently asked questions
Did AMIE replace doctors? No. Physicians supervised the research workflow.
How many patients participated? The reported clinical study included 98 people.
Does this validate AI for independent diagnosis? No; larger controlled evaluations are needed.
Source and further reading
This report is original SenseCentral analysis of the primary announcement or study summary. Dates and reported figures are attributed to the source; interpretations are identified as analysis.