Do AI Scribes Actually Save Doctors Time?

Use of artificial intelligence (AI)-enabled ambient documentation tools, or AI scribes, was associated with modest decreases in time spent in the electronic health record (EHR) and documentation time, a multisite study suggested.

A difference-in-differences analysis of clinicians at five large health systems found that using AI scribes was associated with 13.4 fewer minutes of EHR time (95% CI 9.1-17.7) and 16.0 fewer minutes charting (95% CI 13.7-18.3), reported Lisa Rotenstein, MD, of the University of California San Francisco, and colleagues.

However, using AI scribes did not significantly change the amount of time spent in the EHR outside of work hours, they wrote in JAMA.

“While the time savings are modest, it is likely because there is so much work to do related to patient care that when clinicians were saving time on documentation, they were likely reallocating that to other patient care tasks — messages, chart review, corresponding with other team members, etc.,” Rotenstein told MedPage Today.

Still, this saved time allowed for 0.49 additional weekly patient visits — about one additional patient every 2 weeks — which equated to about $167.37 of extra monthly revenue, according to an exploratory analysis.

Sensitivity analyses yielded similar results. Some groups saw more benefit from AI scribes, including primary care specialists, advanced practice clinicians, female clinicians, and clinicians who used AI scribes in half or more of their visits.

Charting is time-intensive and tied to clinician burnout. Previous research from this study group found that AI scribes reduced burnout and improved well-being. Taken with these new findings, Rotenstein said it “suggests that AI scribes may have a modest benefit for time expenditure but an important impact on how clinicians feel about their time expenditure.”

In an accompanying editorial, Vincent Liu, MD, of the Kaiser Permanente Division of Research in Pleasanton, California, and co-editorialists wrote that this study solidifies that AI scribes can reduce documentation time, though the next question is “whether that time is reinvested in ways that measurably improve outcomes and equity for patients” and how it impacts clinical practice.

Liu and colleagues also noted that soon there will be an “AI-native” generation of clinicians who will have only ever known the field with these tools. They argue that future work should assess these “trainees’ clinical reasoning, documentation quality, skills acquisition, and supervision.”

This study took place across five academic health centers from June 2023 to August 2025. AI scribes were made available to ambulatory clinicians and data were provided from a minimum of 12 weeks, including 6 weeks before AI adoption and 6 weeks after. All used Epic as their EHR vendor and Ambience, Nuance DAX Copilot, and/or Abridge AI scribes.

Rotenstein noted that each health system introduced the AI scribes in their own way and clinicians ultimately chose how they used the tools.

“This means that our findings reflect what might happen upon real world, imperfect introduction of AI scribes in health systems,” she said.

In total, there were 8,581 clinicians of whom 1,809 opted to use AI scribes and 6,772 chose not to; information was collected for both groups. Nearly three-quarters were attending physicians (74.1%), while 18.1% were advanced practice clinicians and 7.8% were resident physicians. Among the cohort, 24.4% were in primary care, 62.4% in medical specialties, and 13.2% in surgical specialties; 57.1% were female.

The primary outcomes were total time spent on the EHR, time spent on documentation, and time spent on the EHR outside scheduled hours, all normalized to 8 scheduled patient hours.

Authors listed several limitations to their work, including the study’s observational design, risk for unobserved confounders, and that implementation details varied by site. Plus, the results may not apply to nonacademic settings, and the analysis did not measure burnout or how time was reallocated. Lastly, the analysis shouldn’t be used for cost-benefit considerations.

Rotenstein and team suggested that future work assess how these tools can best improve clinician’s workflow.