An AI assistant interviewed patients before their doctor's visit, and the first real-world trial found no safety problems
Doctors at Harvard and Boston's Beth Israel Deaconess tested an assistant that took patients' medical histories from home with 98 real patients. No conversation had to be stopped, and most doctors who read the summary said it helped them prepare. It is a small feasibility study.
Before an urgent primary care appointment, 98 patients in Boston talked from home with AMIE, an artificial intelligence assistant built to take a medical history: asking about symptoms, past illnesses and medication, suggesting possible diagnoses and leaving a summary for the doctor. The study, published on 8 October in The Lancet, is the first test of this kind of tool with real patients rather than simulated cases.
How it worked
The trial ran from April to November 2025 at an urgent primary care practice at Beth Israel Deaconess Medical Center (BIDMC), a Harvard teaching hospital. 114 people enrolled and 98 completed both the AI conversation and their appointment. An internist watched every chat live and could step in at any time.
What went well
- No conversation had to be stopped for safety reasons.
- Patients and physician reviewers rated the quality of the conversations favourably on most measures.
- Patients’ attitudes towards AI improved after the chat and stayed that way after the visit.
- In 44 cases the doctor read the transcript or summary beforehand: in 75% they said it helped them prepare.
What it doesn’t show
This is a safety and feasibility study; it does not show that the tool improves anyone’s health. The sample is small and from a single practice. Doctors spotted one made-up error (a past operation dated in the future) and added clarifications in five cases. One clinician rated a conversation as somewhat harmful, because a patient may have been worried to see lymphoma among the possible diagnoses. Patients’ biggest concern was confidentiality.
The study was funded by Alphabet (Google), which develops AMIE, and several authors work for the company. The team led by Adam Rodman, of Harvard and BIDMC, describes it as a necessary step before larger trials.
