A Lancet Study Put an AI Chatbot Ahead of Urgent Primary Care Visits. It Shows Safety in 98 Patients, Not Better Outcomes.
In a Beth Israel Deaconess feasibility study funded by Alphabet, Google's AMIE interviewed patients before visits while a physician watched. No conversation needed a safety stop, but the study did not test whether care improved.
Researchers at Beth Israel Deaconess Medical Center (BIDMC) and Google have reported what BIDMC calls the first prospective real-world study of a patient-facing conversational AI in primary care. The paper, "Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study," was published in The Lancet on Oct. 8, and BIDMC described the results on Oct. 9.
The finding is narrow and useful: under close supervision, the tool did not cause obvious harm in a small group of urgent-care patients. It does not show that pre-visit AI interviews improve care.
How the study worked
The system, Google's Articulate Medical Intelligence Explorer (AMIE), was used from home by patients who had booked an urgent primary care appointment with a BIDMC physician. Through a secure text chat, it asked about symptoms, took a medical history, offered possible diagnoses for patients to review with their doctor and produced a summary for the clinician before the visit. A board-certified internal medicine physician monitored every conversation in real time and could intervene. Enrollment ran from April to November 2025: 114 patients enrolled, and 98 completed both the AI interaction and the visit.
What it found
No conversation required a safety-stop intervention. Supervising physicians identified one hallucination and added clinical clarification in five cases. Primary care physicians reviewed the AI transcript or summary before 44 visits; in 75 percent of those, they said it helped them prepare, and in 57 percent they said it may have influenced their approach. In one case a physician rated the interaction somewhat harmful, out of concern that the patient may have become anxious after AMIE listed lymphoma among the possible diagnoses.
Patients and physician evaluators rated the conversations favorably on most measures, and patients' attitudes toward AI improved after the interaction. Patients were less sure about the confidentiality of what they shared and about the chatbot's honesty and trustworthiness.
Zero safety stops with a physician watching every exchange is a result about supervision as much as software. It does not say what happens when no one is watching.
Limits and conflicts
The investigators say the study was designed to assess feasibility, not to determine whether AI improves health outcomes. It had 98 completers at one academic center and no comparison group. Alphabet funded it, and senior author Adam Rodman, M.D., BIDMC's director of AI programs, was a visiting researcher at Google during part of the study. Rodman said in the release that "AI tools must first be shown to operate safely and acceptably in real care workflows." Co-senior author Marc L. Cohen, M.D., BIDMC's clinical chief of primary care, said the study "serves as an exemplar of how the safety of AI may be assessed prior to clinical implementation."
What to watch
If a vendor pitches you a pre-visit AI interview, ask whether a clinician reviews the exchanges live or after the fact, what the escalation rule is, and where patient answers are stored. Ask too how a list of possible diagnoses shown to a patient is handled, given the anxiety case. For the governance questions, see our piece on what an AI scribe actually changes.
Reported, fact-checked and edited by The Script Pad's editorial staff.
This article is for professional education and does not replace clinical judgment. Treatment decisions should be based on the individual patient and current guidelines.
