Study Finds Google’s AMIE May Help Prepare Patients for Primary-Care Visits

Google’s experimental conversational medical AI AMIE may help organize patient concerns before primary-care appointments, but a small feasibility study did not establish that it improves care or works safely at scale.

AMIE, short for Articulate Medical Intelligence Explorer, conducted text conversations with 100 adults as many as five days before scheduled urgent-care appointments at Beth Israel Deaconess Medical Center in Boston. The interactions took place from April through November 2025 and were followed by primary-care visits.

Ninety-eight patients completed both the AMIE interaction and a provider visit. After chart reviews conducted eight weeks later, the final diagnosis appeared among AMIE’s first seven suggestions in 88 of those 98 cases, or 90%. It appeared among the first three suggestions in 73 cases, or 75%, and was AMIE’s top suggestion in 55 cases, or 56%.

Physicians received AMIE’s transcript and summary before the appointment. In 44 completed surveys, doctors rated the preparation helpful in 75% of cases. Interviews indicated that the material helped shift some visits away from basic information gathering and toward verifying details, counseling patients and making decisions together.

Three independent clinical evaluators compared AMIE’s diagnostic suggestions with those of primary-care physicians. They found similar overall quality, with no significant difference in the appropriateness or safety of the management plans. Physicians rated their own plans higher for practicality and cost-effectiveness, with reported statistical values of p = 0.003 and p = 0.004, respectively.

Human safety supervisors monitored every patient-AMIE interaction in real time under predefined safety-stop criteria. No safety stops were required during the 100 completed interactions. Patients’ attitudes toward AI also improved significantly after using the system, with a reported p-value below 0.001.

The findings suggest AMIE may help as an assistive tool by giving clinicians a structured account of a patient’s concerns before a visit. The study offers no evidence supporting autonomous use of AMIE or substituting it for a clinician.

The study was conducted at one center, enrolled no control group and used continuous physician supervision. It therefore cannot quantify AMIE’s effectiveness compared with the usual pre-visit process. Patients requiring emergency treatment were excluded, as were pregnant patients and those whose main complaint involved mental health.

The results provide an early clinical test of a system previously evaluated mainly through simulated patient interactions. Further research would be needed to assess performance across broader patient groups and settings, including situations without live clinical oversight.