While the patient is still putting on their shoes, a doctor somewhere in a primary care clinic completes a patient visit, opens the electronic health record on a wall-mounted monitor, and begins typing—reconstructing the conversation, the examination results, the plan, the prescriptions, and the follow-up instructions. The AI scribe industry was designed to do away with this ritual, which is performed hundreds of thousands of times a day during clinical encounters in America. or at least to make it much less uncomfortable. The argument was persuasive: let the machine listen, let it write the note, let the doctor check and sign it, and recover the time lost to paperwork that could have been used for patient care.
In April 2026, JAMA published the largest study ever done on the use of AI scribes in actual clinical settings, which followed over 1,800 clinicians from 2023 to 2025 across five academic medical centers in the United States. The findings are truly intriguing. Additionally, they are more modest than the industry’s promotional materials would imply in ways that merit more open discussion than they have received.
| Category | Details |
|---|---|
| Study Title | “Changes in Clinician Time Expenditure and Visit Quantity With Adoption of AI-Powered Scribes: A Multisite Study” — published in JAMA, April 2026 |
| Lead Institutions | Mass General Brigham and University of California, San Francisco (UCSF) — co-led study across five U.S. academic medical centers |
| Study Scale | 1,800+ clinicians using AI scribes compared with 6,770 control clinicians; data collected 2023–2025 |
| Key Finding: Time Saved | AI scribes reduced EHR usage by 13 minutes per day and documentation time by 16 minutes per day — relative decreases of 3% and 10% respectively |
| Productivity Gain | 0.5 additional patient visits per week; one extra patient seen roughly every two weeks per clinician |
| Revenue Impact | Statistically significant but modest — approximately $167 per month per clinician adopting the tool |
| Usage Gap — Critical Finding | Only 32% of users adopted AI scribes in more than 50% of patient visits — the threshold at which benefits doubled or tripled |
| Who Benefitted Most | Primary care physicians, advanced practice providers, and female clinicians saw the most pronounced improvements in documentation burden |
| After-Hours EHR Time | No significant difference in time spent on electronic records outside of working hours — the notorious “pajama time” problem was not measurably resolved |
| Burnout Connection | Prior Yale School of Medicine research (Oct 2025) found AI scribes reduced physician burnout odds by 74%; the JAMA study notes that modest time savings alone cannot fully explain burnout reduction |
| Research Collaborative | First published results from the Ambient Clinical Documentation Collaborative (ACDC) — a multi-organizational research effort tracking real-world AI scribe deployment |
| Lead Authors | Lisa Rotenstein, MD, MBA (UCSF/Brigham and Women’s) and Rebecca G. Mishuris, MD (Chief Health Information Officer, Mass General Brigham) |
For each eight-hour shift of patient care, clinicians who used AI scribes saved 16 minutes of documentation time and 13 minutes in the electronic health record. They saw about half as many patients each week, or one extra visit every two weeks. Gains in revenue, which came to roughly $167 per month per clinician, were statistically significant. These are genuine advancements. However, these aren’t precisely the game-changing figures that hospital systems considering six- and seven-figure technology investments are looking for.
It is worthwhile to consider what remained the same. There was no discernible difference between those who used AI scribes and those who didn’t when it came to after-hours EHR time—the “pajama time” that doctors particularly lament, the notes written at 10 p.m. on a laptop in bed, the charts completed at midnight after a day that was already too long. The technology did not significantly alleviate that specific burden, which is at the core of many physician dissatisfactions and significantly contributes to the kind of grinding exhaustion that ultimately pushes gifted clinicians out of medicine. That finding is worth considering for administrators who purchased AI scribes in part on the promise of lowering burnout by returning doctors’ evenings.
It’s true that the picture of burnout is more nuanced than just time savings. In contrast to the JAMA study’s 16-minute daily time savings, ambient AI scribes reduced physician burnout odds by 74%, according to a Yale study published in October 2025. The JAMA authors are clear about this conflict, pointing out that slight decreases in documentation time are unlikely to completely explain burnout shifts. This implies that something else is going on, perhaps related to the psychological impact of having a tool that manages the cognitive strain of taking notes, even if it doesn’t significantly shorten the time. That’s an important distinction. However, it also makes it more difficult to assess the technology using the simple ROI frameworks that hospital CFOs typically prefer.
The usage gap is, in a sense, the most profound finding. Compared to lighter users, clinicians who employed AI scribes in more than half of their patient visits saw a twofold decrease in overall EHR time and a threefold decrease in documentation time. The relationship between dose and response is evident. Benefit more by using it more. However, only 32% of study participants went over that cutoff. Only a small portion of the tool’s potential benefits were being realized by the remaining 68% of users, who essentially paid for a gym membership, went twice a month, and then wondered why their fitness hadn’t improved. Although the parallel isn’t perfect, the problem’s structure is comparable. Adoption does not produce results if it is not used consistently.
Observing the growth of the AI scribe market gives the impression that the technology and healthcare sectors have somewhat different timelines and success criteria. Adoption curves are measured by tech companies, outcomes per dollar are measured by hospitals, and whether a patient’s day truly improved is measured by clinicians. These metrics aren’t always the same. The AI scribe companies have a right to highlight the productivity gains and burnout reduction data because those numbers are real. Several of these companies have raised significant venture capital based on predictions of widespread clinical adoption. However, the JAMA study adds complexity to a narrative that was being presented a bit too smoothly. It was carried out across five actual health systems with actual clinical workflows and usage patterns.
This does not imply that AI scribes are ineffective. Under the correct circumstances, with adequate adoption rates, they do work in the clinical settings where they are most appropriate; primary care seems to gain more from them than specialties, likely due to the type and volume of documentation these clinicians generate. According to the largest study to date, the technology is a tool rather than a solution, and the difference between what it can do and what it actually accomplishes in practice depends solely on how seriously health systems take its implementation. Purchasing the software and declaring it finished is insufficient. It never is.
