Abstract
IMPORTANCE
Artificial intelligence (AI)–enabled scribes have been proposed to reduce electronic health record (EHR) burden and improve clinician satisfaction. There is limited evidence about their associated results across multiple sites and relative benefits for different clinician groups.
OBJECTIVE
To assess the association of AI scribe adoption with changes in EHR time expenditure and visit volume and how associations vary by clinician characteristics.
DESIGN, SETTING, AND PARTICIPANTS
Multisite, longitudinal cohort study of AI scribe adoption conducted at 5 US academic health care institutions that introduced AI scribes to their clinicians between June 2023 and August 2025. Participants were ambulatory clinicians.
INTERVENTIONS
AI scribe adoption, defined as receiving access to an AI scribe. This was determined by opt-in decisions by eligible physicians at 4 of the 5 sites.
MAIN OUTCOME MEASURES
Total time spent on the EHR, time spent on documentation, and time spent on the EHR outside scheduled hours or on unscheduled days, all normalized to 8 scheduled patient hours; weekly visit volume.
RESULTS
The sample comprised 8581 clinicians, including 1809 AI scribe adopters. Participants were 57.1% female and were split between primary care (24.4%), medical (62.4%), and surgical (13.2%) specialties. Most (74.1%) were attending physicians, with 18.1% advanced practice clinicians and 7.8% resident physicians. In a difference-in-differences analysis, AI scribe adoption was associated with 13.4 (95% CI, 9.1-17.7) fewer minutes of EHR time, 16.0 (95% CI, 13.7-18.3) fewer minutes of documentation time, and 0.49 (95% CI, 0.17-0.81) additional weekly visits delivered. Electronic health record time outside work hours did not change significantly. Changes associated with AI scribe adoption were greatest for primary care specialists, advanced practice clinicians, female clinicians, and clinicians who used AI scribes in 50% or more of visits.
CONCLUSIONS AND RELEVANCE
AI scribe adoption was associated with modest decreases in total EHR time and documentation time and with a modest increase in weekly visit volume.
Documentation in the electronic health record (EHR) requires substantial physician time,1 averaging 2.3 hours per 8 hours of patient care.2 EHR time, particularly when occurring outside work hours, is associated with burnout among physicians.3-5 Further, EHR burden may limit clinical capacity, patient access, and quality of care.3,4
Human scribes can reduce EHR time, but their scalability is constrained by cost and workforce availability.6-10 Artificial intelligence (AI)–enabled ambient documentation tools, or AI scribes, have been introduced as a more scalable alternative. Early studies suggest these tools may reduce EHR burden11-14 and improve clinician satisfaction.15-18 However, findings have varied across sites, practice settings, clinician types, and specialties, and the specific details of a single site’s implementation approach or vendor may limit generalizability. Further, many studies use different constructions of EHR use measures, limiting comparability and making meta-analysis challenging. There is limited literature regarding which clinicians benefit most from AI scribes. Similarly, there is limited evidence on the association of AI scribes with changes in productivity.16-18 Both of these factors are critical to informing health system decision-making on where to invest in and deploy finite resources.
To address these gaps, this study evaluated the association of AI scribe adoption in the context of real-world introductions of these technologies with changes in total EHR time, documentation time, work outside of work, and weekly visit volume across 5 health systems. Analyses also examined whether associations varied by specialty, clinician type, sex, and intensity of AI scribe use.
Methods
Study Setting and Population
The study sample included 5 academic health care institutions (Mass General Brigham, Emory Healthcare, University of California at San Francisco, Yale New Haven Health, and the University of California at Davis) in multiple regions of the United States that introduced AI scribes to their clinicians starting in June 2023. Institutions were selected for the study based on their having made AI scribes available to their ambulatory clinicians, and their willingness and capability to pool data for the study. Sites were asked to provide data for both AI scribe adopters, as well as any clinician who did not adopt AI scribes but practiced in a specialty in which at least 1 clinician adopted the tool. eTable 1 in Supplement 1 displays characteristics of the institutions and their AI scribe rollouts, including how clinicians at individual sites were selected for AI scribe adoption. In general, sites defined the clinicians who were eligible for gaining access to the tool (eg, ranging from no restrictions on eligibility to restrictions based on clinical volume or eligibility based on leadership nomination), and clinicians subsequently opted-in to adoption. All sites used Ambience, Nuance DAX Copilot, or Abridge (individually or in combination) and Epic as their EHR vendor. This study was approved by the University of California at San Francisco institutional review board.
The study population consisted of ambulatory clinicians—attending physicians (including clinical fellows), advanced practice clinicians (nurse practitioners and physician assistants), and resident physicians—who had the option to use an AI scribe during the study period. Each institution provided information on clinicians who received access to AI scribes, regardless of whether they used the tool, as well as the clinicians in the same division or department who did not receive access. Clinicians given access to AI scribes were considered AI scribe adopters, beginning on the date each institution reported that clinician received access.1 Nonadopters were those who did not receive access to AI scribes at that time. The primary analysis included all adopters and nonadopters, while sensitivity analyses explored the association of AI scribe use with the outcomes among clinicians with measured intensity of AI scribe use. Each institution provided data for a minimum of 12 weeks, including 6 preadoption and 6 postadoption weeks for adopters.
Variables
Institutions provided demographic information including clinicians’ role (attending, advanced practice clinician, or resident), sex, specialty (categorized into primary care, medical specialties, or surgical specialty, as specified in eTable 2 in Supplement 1), and the proportion of encounters for which clinicians used an AI scribe (when available; some AI scribe vendors could not provide usage-level data, and it could not be inferred from Epic Signal metadata described below).
This data was linked with month-level EHR metadata derived via Epic’s Signal database, with 3 time-based outcomes normalized to 8 scheduled patient hours19: total time spent on the EHR (EHR-8), time spent on the EHR outside of scheduled hours or on unscheduled days (work outside of work [WOW-8]), and time spent on documentation (DocTime-8). Consistent with variables available in the Epic Signal database, data also included the number of weekly ambulatory visits delivered, the proportion of new vs established patient evaluation and management (E/M) encounters, and the proportion of encounters at each E/M level. Clinicians with fewer than 5 ambulatory visits per month in any study month were excluded because Signal does not calculate complete measures for those clinicians.
Main Analytic Approach
The primary analytic approach was a difference-in-differences framework using multivariable ordinary least-squares models. Assumptions underlying these models are described in the eMethods in Supplement 1. All models controlled for number of weekly visits (except for when visit volume was the model outcome), clinician fixed effects, and month fixed effects. Standard errors were clustered at the clinician level. To evaluate the effect of AI scribes over time and to assess the parallel pretrends assumption, we specified an event study that estimated the dynamic effect in each period before and after adoption, using a Callaway and Sant’Anna estimator.20
Subgroup Analyses
To assess how the association of AI scribes with our main outcomes varied by clinician characteristics, we used the same difference-in-differences regression framework with an interaction term for the clinician characteristic such as specialty, role, sex, and AI scribe use intensity, with 1 model for each characteristic-outcome pairing. D in the eMethods in Supplement 1 for additional details.
Numeric Comparison of Preadoption and Postadoption Trends
While our consideration of both never-adopters and those who may later become adopters substantially increases the size of our control group, it does not facilitate simple descriptive comparisons of preadoption and postadoption trends for adopters compared with control clinicians. As described in the eMethods in Supplement 1, we defined a descriptive post period to facilitate numeric comparison of preadoption and postadoption trends.
Sensitivity Analyses
To ensure results were not driven by a single site, we repeated our difference-in-differences models excluding each site sequentially. Second, we ran our models again specifying our treatment variable to exclude clinicians with unknown use intensity. Third, we repeated our models excluding resident physicians. Fourth, excluded clinicians who never adopted AI scribes, using only preadoption months of eventual AI scribe adopters as control observations. Fifth, we ran our models with time-based outcomes without controlling for weekly visits. Sixth, we used the Callaway and Sant’Anna20 estimator in aggregate in addition to the event study described above to facilitate comparisons with our 2-way fixed-effect results. Last, we assessed heterogenous treatment effects across 5 categories of AI scribe use.
Revenue Analysis
In an exploratory analysis, we estimated marginal E/M revenue associated with AI scribe adoption. We first generated a monthly E/M revenue estimate for each clinician by multiplying the number of billed visits at each E/M service level by the corresponding 2025 Medicare Physician Fee Schedule national payment rate. See the eMethods in Supplement 1 for the relevant equation. We used the difference-in-differences framework described above to estimate the monthly marginal E/M visit revenue associated with general AI scribe adoption and conducted an analysis of heterogeneity of treatment effects as for other outcomes.
Results
Sample Characteristics
The sample included 8581 clinicians, including 1809 clinicians who adopted AI scribes and 6772 clinicians who did not adopt AI scribes over 181 273 clinician-month observations (Table). The majority (74.1%) of the sample consisted of attending physicians, more than one-half (57.1%) were female, and clinicians were split between primary care (24.4%), medical specialties (62.4%), and surgical specialties (13.2%). The distribution of clinicians into individual specialties is shown in eTable 2 in Supplement 1.
Table.
Characteristics of Clinicians Who Did and Did Not Adopt AI Scribes Across 5 Institutions, 2023-2025
| Characteristic | No. (%) | |
|---|---|---|
| AI Scribe adopters (n = 1809) |
AI Scribe nonadopters (n = 6772) |
|
| Institution | ||
| Yale New Haven Health System | 77 (63.6) | 44 (36.4) |
| Emory Healthcare | 453 (53.7) | 390 (46.3) |
| University of California at Davis Health | 342 (43.3) | 447 (56.7) |
| University of California at San Francisco Health | 552 (38.4) | 884 (61.6) |
| Mass General Brigham | 385 (7.1) | 5007 (92.9) |
| Clinician role | ||
| Attending (including fellows) | 1574 (24.7) | 4787 (75.3) |
| Advanced practice clinician | 198 (12.8) | 1352 (87.2) |
| Resident | 37 (5.5) | 633 (94.5) |
| Clinician sex | ||
| Female | 1047 (21.4) | 3853 (78.6) |
| Male | 762 (20.7) | 2919 (79.3) |
| Specialty | ||
| Primary carea | 509 (24.3) | 1583 (75.7) |
| Medical specialtiesa | 1080 (20.2) | 4277 (79.8) |
| Surgical specialtiesa | 220 (19.4) | 912 (80.6) |
| Mean intensity of clinician AI scribe use | ||
| <50% of notes | 780 | NA |
| ≥50% of notes | 577 | NA |
| Unknown | 452 | NA |
Abbreviation: NA, not applicable.
Please refer to eTable 2 in Supplement 1 for delineation of specialties in each of these categories.
Unadjusted EHR Time Expenditure and Visit Volume
Baseline values for EHR time and weekly ambulatory visit volume are shown in Figure 1. Descriptive preadoption and postadoption trends for each outcome using the descriptive postadoption period for each clinician are shown in eFigure 1 in Supplement 1.
Figure 1. Electronic Health Record Time Expenditure and Weekly Ambulatory Visit Volume Among AI Scribe Adopters and Nonadopters, for the Overall Sample.
Unadjusted preadoption and postadoption outcome measures for each group are derived via designation of a descriptive post period for all clinicians in an organization beginning in the month when the first clinician in that organization adopted the artificial intelligence (AI) scribe. Difference-in-difference estimates are derived from a multivariable ordinary least-square regression model that adjusts for clinician fixed effects, time fixed effects, and weekly ambulatory visit volume to estimate the relative impact of AI scribe adoption vs no AI scribe adoption (reference) on each outcome. EHR indicates electronic health record; PSH, patient scheduled hours.
aNot plotted because unit is not compatible with scale.
Difference-in-Differences Estimates of Changes in EHR Time Expenditure and Visit Volume
AI scribe adoption was associated with 13.4 (95% CI, 9.1-17.7) fewer minutes of total EHR-8 time and 16.0 (95% CI, 13.7-18.3) fewer minutes of DocTime-8 (Figure 1). WOW-8 changes (−3.1 [95% CI, – 6.8 to 0.5] minutes) with AI scribe adoption were not statistically significant. AI scribe adoption was associated with delivery of 0.5 (95% CI, 0.2-0.8) additional weekly visits (Figure 1). Temporal trends in each of the outcomes are shown in Figure 2, and the numbers of AI scribe adopter and nonadopter observations available at each time point are reported in eTable 3 in Supplement 1. The majority of preadoption estimates had a 95% CI overlapping 0, providing support that the parallel pretrends assumptions were met.
Figure 2. Difference-in-Differences Models as Event Studies, for EHR-8, DocTime-8, WOW-8, and Weekly Ambulatory Encounters.
Each panel shows the Callaway and Sant’Anna event study estimate of the association between artificial intelligence (AI) scribe adoption and the outcome, relative to the month of adoption (vertical dashed line). Shaded areas indicate 95% CIs. Solid horizontal line at y = 0 indicates no effect. Larger confidence intervals in later postadoption periods reflect lower numbers of observations. The full observation count by AI scribe adoption status for each relative month is available in eTable 3 in Supplement 1. DocTime-8 indicates time spent on documentation; EHR-8, total time spent on the electronic health record; PSH, patient scheduled hours; and WOW-8, time spent on the electronic health record outside of scheduled hours or on unscheduled days.
Similar results were seen in sensitivity analyses that sequentially excluded each study site (eTable 4 in Supplement 1), although significant decreases in WOW-8 time were seen in analyses excluding Mass General Brigham and University of California at Davis. Consistent results were additionally seen in a sensitivity analysis excluding adopters whose utilization intensity was unknown (eTable 5 in Supplement 1) and excluding residents (eTable 6 in Supplement 1). Excluding those who never adopted AI scribes from contributing control data to the difference-in-differences analysis (eTable 7 in Supplement 1) also revealed similar findings, as did models without controls for visit volume (eTable 8 in Supplement 1) and models using the Callaway and Sant’Anna estimator (eTable 9 in Supplement 1. AI scribe adoption was also associated with a decrease in WOW-8 time in the latter analysis.
Difference-in-Differences Estimates by Clinician Group
As shown in Figure 3, Figure 4, and eTable 10 in Supplement 1, the association of AI scribe adoption varied by clinician characteristics and AI scribe use intensity. For example, compared with primary care clinicians who did not adopt AI scribes, those who did spent 25.0 (95% CI, 16.9-33.1) fewer minutes of EHR-8 time (Figure 3A) and 26.9 (95% CI, 22.1-31.7) fewer minutes of DocTime-8 (Figure 3B). Female clinician adopters spent 19.0 (95% CI, 13.3-24.6) minutes less in EHR-8 (Figure 3A), 19.9 (95% CI, 16.7-23.0) minutes less in DocTime-8 (Figure 3B), and 6.2 (95% CI, 1.3-11.1) minutes less in WOW-8 (Figure 4A) compared with nonadopters. Advanced practice clinicians and residents who adopted AI scribes had significantly lower EHR-8, WOW-8, and DocTime-8 and had more weekly visits (Figure 4B).
Figure 3. Electronic Health Record Time Among AI Scribe Adopters and Nonadopters, by Clinician Subgroup.
Times reported in hours per 8 patient scheduled hours (PSH). Reference group in each row is artificial intelligence (AI) scribe nonadopters. Unadjusted preadoption and postadoption outcome measures for each group are derived via designation of a descriptive post period for all clinicians in an organization beginning in the month when the first clinician in that organization adopted the AI scribe. Difference-in-difference estimates are derived from a multivariable ordinary least-square regression model that adjusts for clinician fixed effects, time fixed effects, and weekly ambulatory visit volume, and additionally includes an interaction term between each characteristic (eg, specialty, clinician type, clinician sex, intensity of AI scribe use) and AI scribe adoption to estimate the relative impact of AI scribe adoption vs no AI scribe adoption on each outcome for each specialty, clinician type, clinician sex, and intensity of AI scribe use profile. DocTime-8 indicates time spent on documentation; EHR-8, total time spent on the electronic health record.
Figure 4. Electronic Record Time Outside Scheduled Hours and Weekly Ambulatory Visit Volume Among AI Scribe Adopters and Nonadopters, by Clinician Subgroup.
Encounters indicates encounters per week. Reference group in each row is artificial intelligence (AI) scribe nonadopters. Unadjusted preadoption and postadoption outcomes are derived via designation of a descriptive post period for all clinicians in an organization beginning in the month when the first clinician in that organization adopted the AI scribe. Difference-in-difference estimates are derived from a multivariable ordinary least-square regression model that adjusts for clinician fixed effects, time fixed effects, and weekly ambulatory visit volume, and additionally includes an interaction term between each characteristic (eg, specialty, clinician type, clinician sex, intensity of AI scribe use) and AI scribe adoption to estimate the relative impact of AI scribe adoption vs no AI scribe adoption on each outcome for each specialty, clinician type, clinician sex, and intensity of AI scribe use profile. WOW-8 indicates time spent on the electronic health record outside of scheduled hours or on unscheduled days.
Last, clinicians who used an AI scribe for 50% or more of their visits spent 21.3 (95% CI, 13.9-28.7) minutes less of EHR-8 (Figure 3A), 27.3 (95% CI, 23.1-31.6) minutes less of DocTime-8 (Figure 3B), and 5.5 (95% CI, −0.9 to 12.0) minutes less of WOW-8 (Figure 4A) compared with nonadopters. They delivered 1.0 (95% CI, 0.5-1.6) more visits per week (Figure 4B). The association of AI scribe adoption and changes in EHR time measures and visit volume by more granular utilization groups is shown in eTable 11 in Supplement 1 and revealed incremental reductions in DocTime-8 with increasing levels of AI scribe utilization.
Estimated E/M Revenue Increases With AI Scribe Adoption
AI scribe adoption was associated with an additional $167.37 (95% CI, $86.52-$248.21) per clinician who adopted an AI scribe per month (eFigure 2 and eTable 12 in Supplement 1). Marginal E/M revenue estimates by clinician type and utilization intensity are shown in eTable 12 in Supplement 1.
Discussion
This study in 5 academic medical centers found that adoption of AI scribes was associated with modest reductions in total EHR time and documentation time. AI scribe adopters spent 13 minutes fewer using the EHR in total and 16 minutes fewer on documentation per 8 hours of scheduled patient care, representing 3.0% and 10.0% relative decreases in time spent. Specific groups—including female clinicians, advanced practice clinicians, resident physicians, and primary care and medical specialists—experienced greater desirable changes associated with AI scribe adoption. Clinicians who used the AI scribe 50% or more of the time experienced twice the reduction in total EHR time and 3 times the reduction in documentation time. Last, AI scribe adoption was associated with a 1.7% increase in weekly visit volume, translating to a conservatively estimated additional $167 in monthly E/M visit revenue per clinician.
These multisite findings extend evidence from a growing body of single-site studies regarding the EHR time benefits of AI scribes that overall suggest varying levels of efficacy of the tools. In a single-site randomized clinical trial comparing 2 AI scribe products with a control group, one vendor’s product was associated with a 9.5% reduction in time spent on documentation, while the other’s was not.21 In a separate single-site randomized clinical trial, AI scribe use was associated with significant reductions in total EHR time but not in work outside work after excluding outliers.22 Two other studies in single academic health systems have reported 8.5% lower EHR time per appointment and 15.9% lower time in notes per appointment,12 and 20.4% less time in notes per appointment and 30.0% less after-hours work time per workday,15 respectively, with AI scribe use. In contrast, the present analysis demonstrates that overall EHR time decreased by 3.3% and documentation time decreased by 11.7%, without significant reductions in work outside work. Given their derivation from pooled data from 5 health systems, these estimates are less likely to be biased by site-specific implementation details and represent generalizable results regarding the relative time savings associated with clinicians adopting AI scribes in real-world settings.
It is notable that estimates for documentation time reductions exceeded reductions in total EHR time expenditure and that AI scribe adoption was not ultimately associated with significant changes in work outside of working hours. This is consistent with previous studies12,13 and suggests that while clinicians may save time on documentation with AI scribe use, some of those time savings may be reallocated to other patient care activities, such as reviewing current or prior documentation for accuracy, answering electronic inbox messages from patients, addressing test results, or conducting chart review. While AI scribes may not greatly reduce EHR time, given evidence regarding the burnout benefits of these technologies,16,17 it is possible that clinicians view reallocation of time from documentation as valuable, thus influencing improvements in clinician satisfaction.
The multisite nature of this study allowed for a larger sample size, which enabled well-powered exploratory analyses regarding which groups using AI scribes may see the greatest benefits associated with AI scribe use. Compared with individuals in these groups who did not adopt AI scribes, time savings and in some cases productivity improvements were significantly greater for female clinicians, primary care and medical specialty clinicians, and advanced practice clinicians and residents who adopted AI scribes. Additionally, clinicians who used an AI scribe 50% or more of the time derived greater time and productivity benefits than those who used the AI scribes less. These findings regarding greater relative benefit for female and primary care and medical specialty clinicians are consistent with those seen in prior single-site studies.12,23 Use of AI scribes among residents is an area of ongoing discussion and study,24,25 given the critical nature of documentation for learning and unknown implications for resident learning. Notably, the population of residents who used AI scribes was limited at the institutions in this study; thus, further dedicated investigation is needed, related to both the impact of AI scribes on resident time expenditure and their influence on resident learning. Last, despite disproportionate benefits for those who used AI scribes more than one-half of the time, only about 32% of adopters used their AI scribe that frequently, emphasizing the need robust training and support for adopters.
Last, assuming baseline 2025 Medicare Physician Fee Schedule conversion rates, AI scribe adoption was associated with statistically significant yet numeric ally minimal marginal E/M visit revenue generation of $167 per month. Given that none of the organizations in the study required that clinicians book additional patients to qualify for AI scribe use, this marginal revenue may have derived from increased visits booked in available or nontemplated time or changes in the composition of level-4 and level-5 E/M visits coded facilitated by better documentation, as suggested by prior singlesite evidence.26 These estimates could be interpreted as a conservative lower bound of the financial benefits of AI scribes and could be compared with site-specific implementation costs.
This study has both strengths and limitations. Strengths include the large sample size across multiple institutions, clinician types, and specialties. The quasiexperimental difference-in-differences study design with robust controls helps mitigate some concerns of potential bias, strengthening the validity of the estimates. Assessment of AI scribe adoption (rather than a specific level of AI scribe use) and use of an intention-to-treat analysis approach helps identify a floor of benefit associated with AI scribe use, with greater benefits seen in specific groups.
These strengths are balanced by limitations. First, this was an observational study rather than a randomized clinical trial; thus, characteristics of control clinicians do not perfectly reflect those of AI scribe adopters. While clinician and time fixed effects were used in all models, it cannot be ruled out that the observed associations may be driven by unobserved confounders. Implementation details varied across the study sites; thus, estimates reflect averages and assessments of real-world implementation experiences rather than measurements of maximal efficacy of AI scribes. The study’s focus on academic institutions, where the average weekly visit volume was about 20 encounters per week, may not reflect the experiences of clinicians in nonacademic settings who have higher visit volumes. While marginal revenue analyses generated conservative estimates of E/M revenue gains associated with AI scribe use, costs of AI scribe purchase and implementation vary; thus, the analyses cannot generalize to cost-benefit considerations. The present study did not measure physician burnout or assess reallocation of time across activities associated with AI scribe use. While many of the control observations derived from 1 institution (Mass General Brigham) given data collection specifications, the sensitivity analysis excluding data from Mass General Brigham suggested that inclusion of data from this site only makes the presented estimates more conservative. Due to limitations from a subset of the AI scribe vendors in this study, detailed data on intensity of AI scribe use were missing for some clinicians. Last, the influence of AI scribe adoption was estimated for a limited period after adoption (with the most robust estimates available for up to 5 months after adoption), and the influence of adoption on EHR metrics, productivity, and revenue may differ over time as clinicians become more comfortable with AI scribe use.
Conclusions
This multisite, multivendor, controlled study at 5 academic medical centers found that AI scribe adoption was associated with modest but significant 3.0% relative decreases in total EHR time and 10.0% relative decreases in documentation time, respectively, as well as 1.7% relative increases in visit volume. Greater benefits were observed in specific clinician groups. Given the nonrandomized nature of the study, the observed associations could be due to AI scribe implementation but may also be due in part to unmeasured differences between the adopters and nonadopters studied. Future studies should assess the longevity and reproducibility of these observations, as well as specific workflows and supports that can enhance the benefits of this technology.
Supplementary Material
Key Points.
Question
How does adoption of artificial intelligence (AI) scribes associate with changes in electronic health record (EHR) time expenditure and weekly visit volume?
Results
In 5 academic medical centers, AI scribe adoption was associated with decreases in total EHR and documentation time of 13.4 and 16.0 minutes, respectively, and an increase of 0.49 visits per week. Changes associated with AI scribe adoption were greatest for primary care specialists, advanced practice clinicians, female clinicians, and clinicians who used AI scribes in 50% or more of visits.
Meaning
These findings indicate initial, moderately beneficial associations of AI scribes with changes in key EHR activity measures and visit volume.
Funding/Support:
Dr Rotenstein, Dr Holmgren, Mr Thombley, and Ms Sriram were supported by a grant from the Advancing a Healthier Wisconsin Endowment. Dr Adler-Milstein reports that this work was supported in part by a gift from Ken and Kathy Hao to establish the Impact Monitoring Platform for AI in Clinical Care at the University of California, San Francisco. Ms Iannaccone and Ms Frits were supported by AHRQ grant R01HS029470.
Role of the Funder/Sponsor:
The Advancing a Healthier Wisconsin Endowment; University of California, San Francisco; and AHRQ had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Conflict of Interest Disclosures:
Dr Rotenstein reported receiving grants from FeelBetter Inc; serving on an advisory board for Eko Health and on an AI advisory board for Augmedix Inc; and receiving travel support from FeelBetter Inc outside the submitted work. Dr Holmgren reported receiving grants from the American Medical Association, National Institutes of Health, Assistant Secretary for Technology Policy, Arnold Ventures, and California Health Care Foundation outside the submitted work. Dr Schwamm reported serving as a voluntary member of a client hospital advisory committee for Abridge, an ambient AI clinical documentation company, on behalf of Yale New Haven Health System, and as a content advisor on digital health to the Stroke editorial board. Dr You reported receiving a grant from the National Institutes of Health (T15LM007092) during the conduct of the study. Dr Iannaccone reported receiving grants from the Agency for Healthcare Research and Quality (ARHQ) during the conduct of the study. Dr Frits reported receiving grants from AHRQ during the conduct of the study. Dr Landman reported that his employer, Mass General Brigham, had an institutional investment in Abridge during the conduct of the study and receiving personal fees from Abbott outside the submitted work. Dr Singh reported receiving personal fees from Google for participating on a consumer health advisory panel outside the submitted work. Dr Ehrenfeld reported serving as past president of the American Medical Association, publisher of JAMA and the JAMA Network. Dr Melnick reported receiving grants from the National Institute of Drug Abuse, American Medical Association, and AHRQ outside the submitted work. Dr Bates reported holding equity in ValeraHealth, Clew, MDClone, FeelBetter Inc, Guided Clinical Solutions, and AESOP; receiving personal fees from AESOP, FeelBetter Inc, and Relyens outside the submitted work; and holding a patent(PHC-028654) related to intraoperative clinical decision support licensed to Mass General Brigham. No other disclosures were reported.
Data Sharing Statement:
See Supplement 2.
REFERENCES
- 1.Rotenstein LS, Holmgren AJ, Downing NL, Bates DW. Differences in total and after-hours electronic health record time across ambulatory specialties. JAMA Intern Med. 2021;181(6):863–865. doi: 10.1001/jamainternmed.2021.0256 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Holmgren AJ, Sinsky CA, Rotenstein L, Apathy NC. National comparison of ambulatory physician electronic health record use across specialties. J Gen Intern Med. 2024;39(14):2868–2870. doi: 10.1007/s11606-024-08930-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Adler-Milstein J, Zhao W, Willard-Grace R, Knox M, Grumbach K. Electronic health records and burnout: time spent on the electronic health record after hours and message volume associated with exhaustion but not with cynicism among primary care clinicians. JAm Med Inform Assoc. 2020;27(4): 531–538. doi: 10.1093/jamia/ocz220 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Peccoralo LA, Kaplan CA, Pietrzak RH, Charney DS, Ripp JA. The impact of time spent on the electronic health record after work and of clerical work on burnout among clinical faculty. J Am Med Inform Assoc. 2021;28(5):938–947. doi: 10.1093/jamia/ocaa349 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Rotenstein LS, Hendrix N, Phillips RL, Adler-Milstein J. Team and electronic health record features and burnout among family physicians. JAMA Netw Open. 2024;7(11):e2442687. doi: 10.1001/jamanetworkopen.2024.42687 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Zallman L, Finnegan K, Roll D, Todaro M, Oneiz R, Sayah A. Impact of medical scribes in primary care on productivity, face-to-face time, and patient comfort. J Am Board Fam Med. 2018;31(4):612–619. doi: 10.3122/jabfm.2018.04.170325 [DOI] [PubMed] [Google Scholar]
- 7.Rotenstein L, Melnick ER, Iannaccone C, et al. Virtual scribes and physician time spent on electronic health records. JAMA Netw Open. 2024;7 (5):e2413140. doi: 10.1001/jamanetworkopen.2024.13140 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Mishra P, Kiang JC, Grant RW. Association of medical scribes in primary care with physician workflow and patient experience. JAMA Intern Med. 2018;178(11):1467–1472. doi: 10.1001/jamainternmed.2018.3956 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gidwani R, Nguyen C, Kofoed A, et al. Impact of scribes on physician satisfaction, patient satisfaction, and charting efficiency: a randomized controlled trial. Ann Fam Med. 2017;15(5):427–433. doi: 10.1370/afm.2122 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Micek MA, Arndt B, Tuan WJ, et al. Physician burnout and timing of electronic health record use. ACI open. 2020;4(1):e1–e8. doi: 10.1055/s-0039-3401815 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Ma SP, Liang AS, Shah SJ, et al. Ambient artificial intelligence scribes: utilization and impact on documentation time. JAm Med Inform Assoc. 2025; 32(2):381–385. doi: 10.1093/jamia/ocae304 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Pearlman K, Wan W, Shah S, Laiteerapong N. Use of an AI scribe and electronic health record efficiency. JAMA Netw Open. 2025;8(10):e2537000. doi: 10.1001/jamanetworkopen.2025.37000 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Tierney AA, Gayre G, Hoberman B, et al. Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst. Published February 21, 2024. https://catalyst.nejm.org/doi/full/10.1056/CAT.23.0404 [Google Scholar]
- 14.Shin HS, Williams H, Braykov N, Jahan A, Meller J, Orenstein EW. The influence of artificial intelligence scribes on clinician experience and efficiency among pediatric subspecialists: a rapid, randomized quality improvement trial. Appl Clin Inform. 2025;16(4):1041–1052. doi: 10.1055/a-2657-8087 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Duggan MJ, Gervase J, Schoenbaum A, et al. Clinician experiences with ambient scribe technology to assist with documentation burden and efficiency. JAMA Netw Open. 2025;8(2): e2460637. doi: 10.1001/jamanetworkopen.2024.60637 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Olson KD, Meeker D, Troup M, et al. Use of ambient AI scribes to reduce administrative burden and professional burnout. JAMA Netw Open. 2025; 8(10):e2534976. doi: 10.1001/jamanetworkopen.2025.34976 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.You JG, Dbouk RH, Landman A, et al. Ambient documentation technology in clinician experience of documentation burden and burnout. JAMA Netw Open. 2025;8(8):e2528056. doi: 10.1001/jamanetworkopen.2025.28056 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Shah SJ, Crowell T, Jeong Y, et al. Physician perspectives on ambient AI scribes. JAMA Netw Open. 2025;8(3):e251904. doi: 10.1001/jamanetworkopen.2025.1904 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sinsky CA, Rule A, Cohen G, et al. Metrics for assessing physician activity using electronic health record log data. J Am Med Inform Assoc. 2020;27 (4):639–643. doi: 10.1093/jamia/ocz223 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Callaway B, Sant’Anna PHC. Difference-in-differences with multiple time periods. J Econom. 2021;225(2):200–230. doi: 10.1016/j.jeconom.2020.12.001 [DOI] [Google Scholar]
- 21.Lukac PJ, Turner W, Vangala S, et al. Ambient AI scribes in clinical practice: a randomized trial. NEJM AI. Published online November 26, 2025. doi: 10.1056/AIoa2501000 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Afshar M, Ryan Baumann M, Resnik F, et al. A pragmatic randomized controlled trial of ambient artificial intelligence to improve health practitioner well-being. NEJM AI. Published online November 26, 2025. doi: 10.1056/AIoa2500945 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Stults CD, Deng S, Martinez MC, et al. Evaluation of an ambient artificial intelligence documentation platform for clinicians. JAMA Netw Open. 2025;8(5):e258614. doi: 10.1001/jamanetworkopen.2025.8614 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Chen JL, Tran HN, Brickner LA, Ramalingam ND, Luxenberg A. Noteworthy—integration of ambient AI scribe technology in internal medicine residency clinic. J Gen Intern Med. Published online November 4, 2025. doi: 10.1007/s11606-025-09932-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wright DS, Kanaparthy NS, Melnick ER, et al. The effect of ambient artificial intelligence scribes on trainee documentation burden. Appl Clin Inform. 2025;16(4):872–878. doi: 10.1055/a-2647-1142 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Holmgren AJ, Fenton CL, Thombley R, et al. Ambient artificial intelligence scribes and physician financial productivity. JAMA Netw Open. 2026;9(1): e2553233. doi: 10.1001/jamanetworkopen.2025.53233 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
See Supplement 2.