Google’s conversational diagnostic system, AMIE, has completed its first test with real patients rather than actors. The prospective feasibility study, published in The Lancet on October 8 and assessed by STAT on October 13, ran at Beth Israel Deaconess Medical Center’s primary care practice, where patients with a single urgent complaint chatted with the system from home up to five days before their visit.
The design was deliberately conservative. Of 114 patients enrolled between April and November 2025, 100 completed the AMIE conversation and 98 completed both the chat and their physician appointment. A physician supervisor monitored every conversation live and could intervene under preset criteria. None of the conversations triggered a safety stop. Supervisors noted one hallucination and added clinical information in five interactions. The treating physician received the transcript and a summary before the visit.
On diagnosis, the eventual diagnosis appeared somewhere in AMIE’s differential list in 90 percent of cases and was its top ranked suggestion in 56 percent. Evaluators found no significant difference from physicians on differential quality or management safety, but physicians produced more practical and affordable management plans. Clinicians who reviewed the transcript before the visit, in 44 of 60 surveyed cases, found it helpful for preparation in 33 and said it might have changed their approach in 25. In one case a physician judged the interaction somewhat harmful because a listed possibility, lymphoma, may have made the patient anxious.
The correct reading is narrow. This was a single center, single arm feasibility study with no control group, text only interaction and no access to the medical record, funded by Alphabet with many authors holding Google roles. It shows supervised pre visit history taking by a chatbot can run safely in a real clinic. It does not show better outcomes, saved time or capacity, which are the claims that will decide adoption and payment.