An AI conversation before an appointment can give a doctor a more organised starting point. In a prospective study at Beth Israel Deaconess Medical Center, treating clinicians said material from Google’s AMIE helped them prepare in 33 of 44 cases they reviewed. Patients used the system from home after booking an urgent primary-care visit, answering questions about symptoms and medical history before meeting their doctor.[1][2]
The study enrolled 114 patients between April and November 2025; 98 completed both the conversation and the appointment. It was published in The Lancet on October 8. A board-certified physician monitored each exchange in real time and could step in, so the findings apply to this supervised intake process.[1][2]
What the doctor received
AMIE—Articulate Medical Intelligence Explorer—collected the patient’s account, offered possible diagnoses to discuss with a clinician and prepared a summary. That creates a bridge between the patient’s description at home and the information the doctor needs at the visit. It also gives the patient time to organise details that can otherwise emerge piecemeal during a short appointment.[2]
Among the 44 cases in which treating clinicians reviewed the material, 33 found it helpful for preparation and 25 said it might have influenced their approach. Those are reports of usefulness, rather than a measured reduction in appointment time or an improvement in health outcomes. The study was designed to establish how the conversations functioned in a real clinical setting.[1][3]
The study’s main observations
| Observation | Count | Population |
|---|---|---|
| Completed AI conversation and visit | 98 | 114 enrolled patients |
| Useful for doctor preparation | 33 | 44 clinician-reviewed cases |
| Might have changed the doctor’s approach | 25 | 44 clinician-reviewed cases |
| Predefined safety-stop interventions | 0 | 98 completed encounters |
| Hallucinated surgery date | 1 | Reported by supervisors |
| Additional clinical clarification | 5 | Reported by supervisors |
Patients’ experience of the conversation
Patients generally rated the exchanges favourably, including the system’s listening and explanations. Concerns remained about confidentiality and whether the chatbot could be trusted. Supervisors found one incorrect statement that placed a past operation in the future and supplied clarifications in five cases; none of the encounters met the predefined criteria for stopping the session.[2]
A useful history is more than a list of symptoms. The sequence of events, the patient’s understanding and what they are most worried about can all shape the visit. The opportunity for this kind of tool is to make those details easier to capture and review, with the clinician able to follow up on an answer rather than repeat the whole interview.
The role tested here was preparation: collecting a patient’s account before the visit and giving the treating doctor a summary to review. Clinicians who used that material often found it helpful. Real-time supervision also gave the study a way to identify incorrect information and supply clarifications within the intake process.[1][2]
Sources & further reading
- Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study The Lancet · 8 Oct 2026 · Accessed 25 Oct 2026
- A Real-World Study of Patient-Facing AI in Primary Care Harvard Medical School · 13 Oct 2026 · Accessed 25 Oct 2026
- Study in The Lancet suggests AI could improve patient-physician relationships Google · 8 Oct 2026 · Accessed 25 Oct 2026