Thousands of doctors at The Permanente Medical Group began using software to draft notes from patient conversations in October 2023. Within ten weeks, 3,442 physicians had used it in 303,266 encounters, according to the group’s published evaluation. A smartphone microphone captured conversations for a system that prepared clinical notes for physicians to edit.
Some participating doctors reported that spending less time documenting visits helped them concentrate on patients. The evaluation described favorable early feedback and less time in electronic records, while identifying improvements still needed in medical interpretation and integration with existing software.
A 2025 study in JAMA Network Open examined clinician experience across six health systems. It included 263 doctors and advanced practice clinicians who completed surveys before and after 30 days with an AI scribe. In its adjusted burnout analysis of 184 participants, the proportion reporting burnout fell from 51.9 percent to 38.8 percent, a decline of 13.1 percentage points.
Participants also reported being better able to give patients their undivided attention. That finding came from clinicians’ own ratings. The study had no control group, and early adopters may have been unusually receptive to new technology. Survey responses were not paired with electronic-record measurements of documentation speed.
What computer logs showed
For a separate JAMA study published April 1, 2026, researchers led by Lisa Rotenstein used electronic-record activity from 8,581 clinicians at five academic medical centers. Of those clinicians, 1,809 received access to AI scribes. Researchers compared changes among users and nonusers rather than randomly assigning the tools.
Scribe adoption was associated with 16 fewer minutes documenting care and 13.4 fewer minutes in electronic records overall, per eight scheduled patient-care hours. After-hours electronic-record time did not fall significantly. Documentation was one part of a workday that also included messages, test results and chart review.
Patient consent and clinician review were explicit parts of Kaiser Permanente’s August 2024 rollout of Abridge, an AI documentation product, across 40 hospitals and more than 600 medical offices. Kaiser said doctors and other clinicians would review drafts before placing them in medical records. Its announcement followed testing and quality checks and described software that summarized spoken medical information.
That procedure leaves a doctor responsible for deciding whether a draft accurately describes a visit. A useful evaluation would therefore examine the signed note after corrections, as well as the draft: what was omitted, what was added, and how much editing was required.
A question for medical educators
The studies address several different experiences: time at a computer, perceived workload and attention during visits. None of those measurements directly tests whether a doctor has become better or worse at interpreting symptoms. A study of that question would need to assess clinical reasoning over time, including performance when the scribe is unavailable.
Rotenstein and colleagues specifically identified resident learning as an area needing further investigation. Their sample included 37 resident physicians with access to AI scribes, compared with 633 residents without access. Dedicated research on learning would need to assess outcomes beyond computer time and visit counts.
