What is an AI medical scribe?

How it works

An AI scribe, often called an ambient scribe, captures an appointment with little effort from the clinician. The 29 July 2026 version of NHS England guidance lists the typical functions:

• capturing speech
• converting the recording into a transcript
• generating a summary
• placing the result in the health record

The latest products use generative AI, it says, and some also link terms to clinical codes or draft referral and patient letters. Users should review and approve outputs before further action, the guidance adds.

An American Medical Association (AMA) article says The Permanente Medical Group's scribes draft notes for physicians to edit. It says they offer no diagnoses or treatment suggestions, unlike clinical decision support tools.

What the evidence shows

Two randomized trials were published in NEJM AI on November 26, 2025. One assigned 238 outpatient physicians at a single academic institution to Microsoft's DAX Copilot, Nabla or usual care from November 2024 to January 2025. Nabla users had a 9.5% decrease in note-writing time relative to the control group. DAX users showed no significant change. Both tools were linked to possible improvements in burnout scores. The authors said these need confirmation in larger trials. Physicians reported occasional clinically significant inaccuracies.

The other was a 24-week trial of 66 practitioners at clinics in two states. It found lower work exhaustion and 0.36 fewer hours per day spent on notes, but no significant gain in professional fulfillment.

The AMA's June 2025 account of The Permanente Medical Group's NEJM Catalyst analysis gives routine-use figures. The group reported that 7,260 of its physicians used AI scribes in about 2.58 million patient encounters between October 2023 and December 2024. That saved an estimated 15,791 hours of documentation time, it said.

Where things stand in 2026

A Veterans Health Administration study was published in Annals of Internal Medicine in 2026. In it, 11 AI scribe tools and 18 human note takers wrote up five standardized primary care cases. Blinded raters scored the human notes higher in all five. The cases were simulated. The human notes were not written under real-world constraints. The authors called for independent evaluations, not tied to vendors, before wide deployment.

In England, the NHS guidance says whether an ambient scribe qualifies as a medical device depends on the intended purpose its manufacturer defines. It lists missing or incorrect information among possible clinical hazards.