AI scribes cut documentation time by 69% in simulated primary care visits, study finds

Primary care physicians in a University of Toronto simulation spent 69.1% less time on documentation when an artificial intelligence (AI) scribe drafted their notes than when they charted manually, according to a study published in July in JAMIA Open.

Documentation took up 36.3% of each encounter without a scribe and 11.2% with one. The visits themselves were not significantly shorter.

Once the conversation ended, physicians working with a scribe spent 23.2% of the encounter finishing the chart, compared with 16.2% without one. The authors said that extra time could reflect reviewing and editing the AI-generated note. That difference narrowly missed statistical significance.

LaShawn Murray, M.S., a doctoral candidate in the university’s Department of Mechanical and Industrial Engineering, led the research, which the university highlighted in a news release this month.

Co-authors include Enid Montague, Ph.D., a University of Toronto professor, and Onil Bhattacharyya, M.D., Ph.D., of Women’s College Hospital in Toronto.

How did the simulation work?

Nine primary care physicians licensed in Ontario each saw four standardized patients, played by members of the research team. The visits took place in a room at Women’s College Hospital’s Virtual Care Lab set up to resemble a primary care office.

The cases, two simple and two complex, included wrist pain after an osteoporosis diagnosis and a follow-up three months after a heart attack.

For one simple case and one complex case, physicians typed their notes into a test version of OSCAR Pro, an electronic medical record commonly used in Ontario primary care.

For the other two, an AI scribe drafted the note, which the physician reviewed, pasted into the record and edited as needed. The scribes were web-based tools that did not integrate with the record, and the paper does not name them.

Two researchers coded video of every visit, timing typing, scrolling, handwriting and copy-and-paste work to the second. Six of the nine physicians had used an AI scribe before. The Ontario Ministry of Health and the Natural Sciences and Engineering Research Council of Canada funded the study, and the authors reported no conflicts of interest.

Where did the saved minutes go?

Documentation averaged about 5 1/2 minutes per visit without a scribe and less than two minutes with one. Scribe-assisted visits ran about a minute shorter on average, a gap that was not statistically significant.

Scrolling took up more of each scribe-assisted visit, 4.6% compared with 1.2%. The paper does not report how physicians spent the time freed from typing.

“We thought it would save five minutes a visit or 10 minutes a visit. It turns out, it doesn’t save that much time,” said Robert Wachter, M.D., professor and chair of the Department of Medicine at the University of California, San Francisco (UCSF), in an interview with Medical Economics earlier this year.

“It probably does save me a decent amount of time typing and documenting, and some of that time is repurposed into actual, genuine human contact, which feels better for the doctor and feels better for the patient,” Wachter said.

“Some clinicians are using documentation AI to their advantage, but it depends on how their practice works,” said Sarah Matt, M.D., MBA, a physician executive who hosts Medical Economics’ “Vital Signs,” in an interview with Medical Economics in August. “Are they now shoving more patients into their schedule, or are they going home early?”

“If I save five minutes, do I repurpose that into five minutes of the hamster wheel spinning that much faster?” Wachter said, calling it “a tricky management question.” If part of the goal is a happier physician and patient, he said, “probably you shouldn’t.”

How do lab results compare with everyday practice?

A multisite study published April 1 in JAMA compared 1,809 clinicians who adopted AI scribes with 6,772 who did not at five academic health systems. Adopters spent about 16 fewer minutes on documentation per eight hours of scheduled patient care.

That was a 10% relative drop, and adopters also gained about half a patient visit per week, according to Mass General Brigham, which co-led the study with UCSF. After-hours time in the electronic health record did not differ significantly between the groups.

Gains were largest among primary care physicians, advanced practice providers, female clinicians and those who used a scribe in more than half their visits. Only 32% of users reached that level, and they saw three times the reduction in documentation time.

“We also have some data suggesting that the clinicians who have a relatively greater burden of documentation at baseline benefit proportionally more from ambient documentation technology,” said Lisa Rotenstein, M.D., MBA, M.Sc., a primary care physician and associate professor at UCSF who led the JAMA study. She spoke on an episode of Medical Economics’ “Off the Chart” podcast.

Rebecca G. Mishuris, M.D., M.S., M.P.H., chief health information officer at Mass General Brigham and the study’s senior author, said in a statement that the time reductions “are unlikely to fully account for changes in burnout.” Earlier research found burnout among Mass General Brigham scribe users fell from 52.6% to 30.7% after 84 days.

A flash poll of 103 physicians was presented during an April webinar hosted by Medical Economics and Physicians Practice and sponsored by Heidi Health. In it, 42% said an AI scribe had saved them zero minutes on a typical clinic day, while 20% said it saved more than an hour.

The Toronto paper’s limitations note that physicians had to finish all documentation in the exam room before the next visit and could not use dot phrases or auto text, keyboard shortcuts that save documentation time.

Does writing the note help physicians think?

“The most interesting finding for me was seeing how physically creating the charts throughout the appointment prompted thinking and more clinical decision-making questions from the doctors,” Murray said in the university’s release. “When the scribes were in use, we saw the adoption of scratch pads for note taking to help physicians remember to ask follow-up questions.”

The study protocol allowed physicians to jot notes by hand during scribe-assisted visits to aid recall. Handwriting accounted for 0.2% of those visits’ time on average.

Wachter recalled typing notes and realizing he had forgotten to ask a patient whether she had a pet at home. “There’s something in the process of the human thinking about something and writing it that is probably more cognitively active than the human reading over a draft of something,” he said.

He added that he had not “seen a really good study that kind of gets at this instinct I have that there’s a little bit lost when I’m not typing my own note.”

“Note taking doesn’t add value,” said Josh Umbehr, M.D., a family physician and co-founder of Atlas MD, a direct primary care practice in Wichita, Kansas, in an interview with Medical Economics. “The hourlong conversation with the patient, where you’re keyed in and not typing and not trying to juggle what to remember and what not to, that’s high value.”

Who answers for what the scribe writes?

In the Toronto protocol, a scribe-assisted visit ended only when the physician had reviewed, pasted, edited and saved the note. The paper reports only on documentation time. The authors said future work will cover the quality of the notes.

A January 2025 study in the Journal of Medical Internet Research tested two commercial scribes in simulated visits. It found errors in 70% of the notes, and omissions were the most common type.

“I think it comes back to the standard of care and just ensuring that providers, even if they’re relying heavily on it, that they’re still checking to make sure that whatever is in the medical record is accurate,” said Dan Silverboard, J.D., a health care attorney with Holland & Knight, in an interview with Medical Economics.

“Many of the folks that are using it on the front lines, doctors and nurses, you name it, had nothing to do with the decision to bring that tool into their day-to-day workflow,” Matt said.

Even so, when clinicians document something, “whether it was written by themselves, their resident, their nurse practitioner or by AI, they still signed it, and so it’s still their responsibility,” she said.

What comes after the scribe?

“When the technology was new, the conversation was very much that this would solve all our problems,” Murray said in the release. The conversation has since shifted to viewing scribes as “one piece of the overall documentation burden,” Murray said.

Referrals, prescription orders, billing and regulatory expectations also add to physicians’ administrative work, according to Murray.

“I do think we have to move beyond just documentation into the rest of the work of medicine that doesn’t feel like the work of medicine,” Rotenstein said. AI-drafted replies to patient messages “haven’t been quite as successful,” she said, and she named prior authorization as “a next frontier.”

“If we’re using AI and not telling them, that seems a little iffy,” Matt said of patients.

Murray also wants to know whether patients will hesitate over being recorded and over who has access to the recordings.

“What happens when physicians who are almost always using AI scribes suddenly have one patient that doesn’t consent to it being used in their care?” Murray said.