Google study in The Lancet tests AMIE medical AI before patients see a doctor, finding high alignment with clinicians

Google researchers and clinicians at Beth Israel Deaconess Medical Center (BIDMC) reported a real-world clinical evaluation of AMIE, a medical AI chatbot designed to assess patient needs before an appointment. The study, published in The Lancet, describes how AMIE interacted with patients in a busy ambulatory primary care setting, producing summaries that doctors said helped them prepare and guiding clinician thinking without requiring the conversations to be paused for safety reasons.

The trial took place at BIDMC’s ambulatory primary care clinic, where 98 patients consulted AMIE ahead of urgent care visits. Supervising physicians monitored the interactions in real time and applied predefined safety criteria. According to the report, not a single AMIE conversation needed to be interrupted because of those safety thresholds.

The study also focused on whether AMIE summaries were useful to the clinicians in the moments leading up to the visit. In 75% of cases, clinicians reported that the AI-generated summaries helped them prepare for the appointment. The researchers said AMIE influenced clinicians’ approach to care in more than half of the encounters, suggesting that the system’s outputs were more than background information.

Beyond workflow and clinician reception, the report addresses diagnostic consistency. The study says AMIE’s differential diagnoses, meaning a ranked set of possible conditions it considers based on patient inputs, matched the clinicians’ final diagnoses 90% of the time. The result is framed as an early indicator that AI-generated medical reasoning, at least in this limited setting and protocol, can align closely with physician conclusions.

AMIE, as described in the study write-up, is Google’s research diagnostic AI chatbot. In practice, the system was used as a pre-visit “entry point” to structure patient-reported concerns before clinicians saw the patient, with the explicit goal of helping doctors gather relevant information more efficiently while maintaining safety controls.

For Google, the paper represents a milestone, with the company noting it is the first publication in the main journal of The Lancet for Google. The company positioned the study as a step toward broader use of AI in health settings, particularly in primary care where time constraints can strain both patient communication and clinician attention.

Health systems have been wrestling with how to incorporate AI without eroding trust between patients and clinicians. The report’s emphasis on the patient-physician relationship reflects that concern. Google said the findings suggest the potential for AI to strengthen that interaction, while also easing workload pressures on healthcare workers, though it acknowledged that larger studies will be required to confirm the benefits at scale.

Still, the evidence disclosed so far is limited in scope, and the write-up does not provide details that would be central to evaluating broader readiness, such as the duration of the clinic visits, patient demographics, how outcomes were measured beyond clinician agreement, or what specific categories of safety criteria were applied. The company also said larger clinical trials are needed to assess patient-facing AI at scale, indicating that the current work should be read as an initial, controlled look rather than proof of effectiveness across diverse settings.

What to watch next is whether subsequent studies expand beyond a single clinic and evaluate patient outcomes, safety performance across a wider range of conditions, and how AI summaries affect follow-up decisions over time. The same matters for adoption, including how clinicians and patients respond when AI inputs conflict with initial impressions, and whether the approach remains robust as systems encounter more varied medical presentations.