Ambient AI Tool Adoption in US Hospitals and Associated Factors

ABSTRACT

Objectives: To estimate the prevalence of ambient artificial intelligence (AI) documentation tool adoption among US hospitals using Epic electronic health record (EHR) systems and to identify hospital characteristics associated with adoption.​

Study Design: Cross-sectional observational study of US hospitals using Epic.​

Methods: Among a national sample of US hospitals using Epic, we assessed ambient AI adoption using Epic Showroom (June 2025) to identify eligible ambient applications and health systems that had implemented or were implementing these applications. We linked adoption data to hospital characteristics from the American Hospital Association Annual Survey (2012-2023; most recent response per hospital) and estimated multivariable logistic regression models with robust SEs clustered at the domain level, reporting adjusted predicted probabilities (margins).​

Results: Among 6561 US hospitals, 2784 (42.4%) were Epic users. Among Epic hospitals, 62.6% adopted ambient AI. In adjusted analyses, adoption was higher across workload quartiles (61.7% in quartile [Q] 1 vs 73.1% in Q4; P = .003) and among hospitals in the top operating margin quartiles (58.0% in Q1 vs 67.6% in Q4; P = .001 vs Q1). Adoption was higher among metropolitan hospitals (64.7% vs 54.3% in nonmetropolitan hospitals; P = .012) and nonprofit hospitals (70.2% vs 28.8% in for-profit hospitals; P < .001).​

Conclusions: Ambient AI documentation tools were widely adopted among US hospitals using Epic EHR systems, with adoption associated with workload, financial performance, ownership, and select structural characteristics. These patterns suggest potential for uneven diffusion across hospitals and underscore the need for research on impacts on clinician outcomes, care quality, and equity.

Am J Manag Care. 2026;32(1):e25-e30

Takeaway Points

• Ambient artificial intelligence tools that capture patient-clinician conversations to support documentation were used by nearly two-thirds of US hospitals on Epic electronic health record systems in 2025.
• In adjusted analyses, adoption was higher among hospitals with stronger operating margins, larger size, metropolitan location, nonprofit ownership, and higher staffing-adjusted workload, with geographic variation (lower uptake in the Midwest vs the South).
• Future studies should evaluate impacts on clinician burnout, care quality, and costs, and assess whether uneven diffusion could widen disparities without targeted support for resource-constrained hospitals.

Electronic health records (EHRs) have been widely adopted by hospitals across the US.¹ Clinicians often spend a significant amount of time writing and editing clinical notes within the EHR, with prior estimates suggesting nearly 2 hours of documentation for every hour of direct clinical time.² Documentation burden contributes to burnout,^3-5 which can decrease physicians’ productivity and increase the likelihood of medical errors.⁶ Researchers and practitioners have explored several approaches to reduce documentation burden,⁷ including medical scribes, technology-based solutions (eg, dictation tools), and workflow optimization. Recently, ambient artificial intelligence (AI) technologies have emerged as a potential solution to lessen documentation burden.⁸ These tools capture clinician-patient conversations and generate draft clinical notes for clinician review. Preliminary study findings suggest that these tools may reduce documentation burden and improve patient-provider communication, indicating potential to mitigate physician burnout.^9,10

Little is known about the prevalence of ambient AI across US hospitals or the factors driving its adoption. Prior research has shown that hospitals with certain characteristics (eg, larger size, higher profit margins) are more likely to adopt new health information technologies,¹¹ which could contribute to widening gaps in high-quality care.¹² Existing studies of ambient AI adoption have largely focused on single health systems,^9,10 limiting generalizability to national adoption patterns. A more recent study examining the use of ambient AI among a larger sample of health systems found that all responding systems reported using ambient AI for clinical documentation, which limited the ability to assess adoption disparities.¹³

Using data from Epic on hospitals’ adoption of ambient AI applications, we aim to provide a high-level overview of ambient AI adoption among US hospitals using an EHR system from Epic, the largest EHR vendor in the country.¹⁴ Additionally, we linked hospital-level ambient AI adoption data with American Hospital Association (AHA) Annual Survey data¹⁵ to assess how organizational characteristics are associated with ambient AI adoption patterns.

METHODS

Study Design and Data Sources

We conducted a cross-sectional study of US hospitals to examine hospital characteristics associated with adoption of ambient AI documentation tools as of June 2025 among hospitals using the Epic EHR system. The hospital sampling frame was derived from the June 2025 CMS Hospital Enrollments (“All Hospital Enrollments”) data set, which includes hospitals enrolled in Medicare. We deduplicated records using National Provider Identifier and address matching, yielding 6561 unique hospitals from 9244 enrollment records.

Identification of Epic Hospitals

We used the AHA Annual Survey Information Technology (IT) Supplement (2012-2023) to identify hospitals reporting Epic as their inpatient EHR vendor, using the most recent available survey response for each hospital. Because EHR vendor information was unavailable for a subset of hospitals in the AHA data, we supplemented AHA identification using a domain-level matching approach. Specifically, for hospitals with missing EHR vendor information and for hospitals reported as using a non-Epic EHR, we identified each hospital’s official website and mapped hospitals to website domains; we then matched these domains to Epic’s June 2025 customer list to identify additional hospitals using Epic as their primary inpatient EHR. Using this combined approach, we classified 2784 of 6561 hospitals (42.4%) as Epic inpatient EHR users as of June 2025.

Identification of Ambient AI Adoption

We used Epic Showroom, Epic’s marketplace for third-party applications, to identify ambient AI products. Epic Showroom listed 12 applications under the “Ambient Voice Recognition” category in June 2025. With Epic’s permission, we obtained information on Epic customer health systems that had implemented or were in the process of implementing at least 1 of these applications. Hospitals were classified as having adopted ambient AI if they belonged to an adopting health system, using the same domain-based health system matching approach described earlier.

Predictors of Adoption

We selected potential predictors of ambient AI adoption based on hypotheses related to documentation burden, staffing-adjusted workload, financial capacity, and structural hospital characteristics.

To measure workload intensity relative to staffing, similar to a prior study,¹⁶ we constructed a staffing-adjusted workload measure defined as the sum of each hospital’s annual outpatient visit volume and annual inpatient days divided by total hospital staffing (full-time equivalent [FTE] employees). Annual inpatient days and outpatient visit volume were obtained from the AHA Annual Survey. We set implausibly high outpatient visit totals (> 350,000; n = 3) and implausible inpatient days values (n = 13) to missing and retained these hospitals in analyses using a missing-indicator category. Workload was categorized into quartiles based on its distribution among hospitals with nonmissing numerator and denominator values (Q1-Q4), with Q1 as the reference.

Additionally, following prior studies,^11,12 we hypothesize that financial constraints represent a major barrier to adopting AI technologies. To measure financial performance, we constructed operating margin using each hospital’s most recent AHA Annual Survey response, defined as operating income divided by operating revenue (percentage). We set implausible operating margin values (< −200% or > 200%) to missing (n = 13) and retained these hospitals using a missing-indicator category. Operating margin was categorized into quartiles using nonmissing values (Q1-Q4), with Q1 as the reference.

Lastly, in line with prior studies,^12,17 additional hospital characteristics included hospital size (staffed beds categorized as small [<100], medium [100-399], and large [400-3018]), teaching status, metropolitan location, ownership type (nonprofit, for-profit, or government-owned), Census region (Northeast, Midwest, South, West), case mix index, Disproportionate Share Hospital (DSH) status, and Critical Access Hospital designation.

Statistical Analyses

We first conducted unadjusted descriptive analyses comparing ambient AI adoption across hospital characteristics. Continuous variables are reported as median (IQR) and compared using the Wilcoxon rank-sum test, whereas categorical variables are reported as n (%) and compared using χ² tests.

We then estimated multivariable logistic regression models with ambient AI adoption as the binary outcome and included staffing-adjusted workload quartiles, operating margin quartiles, hospital size, teaching status, metropolitan location, ownership type, Census region, and DSH designation as covariates. Robust SEs were clustered at the domain level. For each categorical predictor, we calculated adjusted predicted probabilities of adoption from the fitted model.

To evaluate robustness to missing covariate handling and potential collinearity induced by missing-indicator categories, we conducted sensitivity analyses reestimating models after excluding hospitals with missing values and compared effect estimates and statistical significance across specifications. Variance inflation factors were examined to assess multicollinearity in primary and sensitivity models. All analyses were conducted using Stata 18 (StataCorp LLC). This study did not involve patient-level data and was deemed exempt from institutional review board review because it used only aggregated hospital-level data.

RESULTS

We estimate that 2784 of 6561 US hospitals (42.4%) used the Epic EHR system in June 2025. Among these, 1744 (62.6%) hospitals had adopted an ambient AI tool, whereas 1040 had not. The 3 most commonly adopted tools are DAX Copilot, Abridge, and ThinkAndor, which together account for more than 80% of all ambient AI implementations among hospitals using Epic.

Hospitals that adopted ambient AI differed from nonadopters across workload, volume, financial performance, and structural characteristics (Table). Adopting hospitals had higher staffing-adjusted workload compared with nonadopters (median total volume per FTE, 208.0 vs 184.0; P < .001) and higher outpatient volume (median annual outpatient visits, 131,405 vs 82,221; P < .001). Adopting hospitals also had higher inpatient volume (median annual inpatient days, 31,609 vs 17,706; P < .001) and positive operating margins (median, 0.9% vs −1.7%; P = .020). Adoption was more common among larger hospitals (12.9% of adopters vs 7.9% of nonadopters were large; P < .001) and metropolitan hospitals (80.9% vs 77.4%; P = .027). Ownership patterns differed substantially: Adopting hospitals were more likely to be nonprofit (87.4% vs 61.4%) and less likely to be for profit (5.5% vs 22.9%) or government owned (7.1% vs 15.7%; P < .001). Region also varied by adoption status, with adopters being less frequently located in the Midwest (28.4% vs 39.0%) and more commonly located in the South (37.5% vs 33.7%) or West (18.7% vs 15.0%; P < .001). Finally, adopting hospitals were more likely to be DSH (46.8% vs 42.5%; P < .001).

Adjusted predicted probabilities of ambient AI adoption are shown in the Figure, and adjusted ORs are provided in eAppendix Table 1 (eAppendix available at ajmc.com). After multivariable adjustment, adoption increased monotonically across workload quartiles (61.7% in Q1 vs 73.1% in Q4; P = .003 for Q4 vs Q1). In contrast, adoption did not differ significantly across outpatient visit or inpatient day quartiles after adjustment. Adoption was also higher among hospitals with stronger operating margins, with higher probabilities in Q3 (68.7%; P = .001) and Q4 (67.6%; P = .009) compared with Q1 (58.0%). Metropolitan hospitals had higher adjusted adoption probability than nonmetropolitan hospitals (64.7% vs 54.3%; P = .012).

Marked differences persisted by ownership and region: Nonprofit hospitals had higher adoption probability than government hospitals (70.2% vs 45.0%; P < .001), whereas for-profit hospitals had lower adoption probability (28.8%; P = .011 vs government). Relative to hospitals in the South (69.5%), adoption rates were lower in the Midwest (54.9%; P = .005), with no statistically significant differences observed for the Northeast or West. DSH status was associated with a higher adoption probability than non-DSH hospitals (64.3% vs 57.8%; P = .038), whereas teaching status was not associated with adoption. Results were consistent in sensitivity analyses excluding hospitals with missing values (eAppendix Table 2).

DISCUSSION

Our analysis shows that ambient AI adoption is widespread, with nearly two-thirds of US hospitals on Epic EHR systems using ambient AI tools. Ambient AI adoption patterns are consistent with findings of prior national studies on hospital adoption of AI technologies and other advanced health IT tools,^11,12 which have shown that factors such as hospital size, patient population, geographic location, operating margins, and ownership status can significantly influence adoption. In adjusted analyses, adoption was higher among hospitals with greater staffing-adjusted workload and stronger operating margins, consistent with the hypothesis that both perceived documentation burden relief and available financial capacity shape adoption decisions.

Additionally, adoption also differed by ownership and geography. Nonprofit hospitals had substantially higher adjusted adoption probabilities than government hospitals, whereas for-profit hospitals had lower adjusted adoption probabilities. These differences may reflect variation in capital availability, organizational priorities, and tolerance for near-term implementation costs when the financial return is uncertain.¹⁸ Geographic differences persisted after adjustment, with hospitals in the Midwest exhibiting lower adoption probabilities than hospitals in the South, suggesting that market factors and vendor/health system penetration may contribute to regional variation.

Most existing evidence on ambient AI has focused on short-term, clinician-centered outcomes such as documentation time, after-hours work, and perceived burden.^9,10 Additional studies are needed to evaluate downstream effects on care processes and patient outcomes, as well as to identify potential unintended consequences (eg, documentation quality, equity, safety). If ambient AI improves clinician efficiency and care quality, uneven adoption could contribute to widening differences in performance and outcomes across hospitals.¹⁹

Our findings also highlight potential policy and market levers to support equitable adoption. Because adoption was higher among hospitals with stronger operating margins, cost and uncertainty about return on investment may be important barriers for financially constrained hospitals. Evidence on cost-effectiveness and implementation supports (eg, shared services, scalable pricing models, technical assistance) could help reduce barriers. Analogous to prior federal efforts that accelerated EHR adoption through financial incentives,²⁰ policy makers could consider targeted approaches to encourage adoption of effective AI technologies while monitoring for unintended increases in disparities.

Limitations

This study has several limitations. First, analyses were limited to US hospitals using Epic as their EHR, which may limit generalizability to hospitals using other vendors with different integration pathways, interoperability constraints, and pricing models. Second, this was an observational, descriptive analysis; residual confounding is possible, and results should be interpreted as associations rather than causal effects. Third, our identification of Epic hospitals and ambient AI adoption relied partly on domain-level matching and health system–level adoption status, which may misclassify some hospitals (eg, mismatched domains, mixed-vendor systems within a health system, partial rollout across facilities). Fourth, some hospital characteristics were drawn from AHA responses through 2023 and may not fully reflect hospitals’ 2025 characteristics. Lastly, several covariates had missing values; although we used missing-indicator categories in primary models to preserve sample size, findings were similar in complete-case sensitivity analyses (eAppendix Table 2), reducing concern that missingness materially altered conclusions.

CONCLUSIONS

Among US hospitals using Epic as their primary inpatient EHR, ambient AI adoption was common in 2025, indicating rapid diffusion of AI-enabled documentation support. In adjusted analyses, adoption was higher among nonprofit hospitals and hospitals with stronger operating margins and was also associated with higher staffing-adjusted workload and select structural characteristics (eg, larger size). Geographic variation persisted after adjustment, with lower adoption in the Midwest relative to the South. Future work should identify modifiable barriers and facilitators to adoption and use of ambient AI, and evaluate downstream impacts of ambient AI use on care quality and patient outcomes. Policy efforts may also be needed to ensure that adoption of new AI technologies does not widen care quality disparities, particularly in smaller, resource-constrained hospitals.

Author Affiliations: Department of Health Policy and Management, Rollins School of Public Health, Emory University (FY, IG), Atlanta, GA.

Source of Funding: None.

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (FY); acquisition of data (FY); analysis and interpretation of data (FY, IG); drafting of the manuscript (FY, IG); critical revision of the manuscript for important intellectual content (FY, IG); statistical analysis (FY, IG); administrative, technical, or logistic support (FY, IG); and supervision (IG).

Address Correspondence to: Ilana Graetz, PhD, Department of Health Policy and Management, Rollins School of Public Health, Emory University, 1518 Clifton Rd NE, Atlanta, GA 30322. Email: ilana.graetz@emory.edu.

REFERENCES

1. National trends in hospital and physician adoption of electronic health records. Assistant Secretary for Technology Policy. Accessed July 1, 2025. https://www.healthit.gov/data/quickstats/national-trends-hospital-and-physician-adoption-electronic-health-records

2. Sinsky C, Colligan L, Li L, et al. Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties. Ann Intern Med. 2016;165(11):753-760. doi:10.7326/m16-0961

3. Downing NL, Bates DW, Longhurst CA. Physician burnout in the electronic health record era: are we ignoring the real cause?. Ann Intern Med. 2018;169(1):50-51. doi:10.7326/m18-0139

4. Raney L, McManaman J, Elsaid M, et al. Multisite quality improvement initiative to repair incomplete electronic medical record documentation as one of many causes of provider burnout. JCO Oncol Pract. 2020;16(11):e1412-e1416. doi:10.1200/op.20.00294

5. West CP, Dyrbye LN, Shanafelt TD. Physician burnout: contributors, consequences and solutions. J Intern Med. 2018;283(6):516-529. doi:10.1111/joim.12752

6. Dyrbye LN, Shanafelt TD. Physician burnout: a potential threat to successful health care reform. JAMA. 2011;305(19):2009-2010. doi:10.1001/jama.2011.652

7. Sloss EA, Abdul S, Aboagyewah MA, et al. Toward alleviating clinician documentation burden: a scoping review of burden reduction efforts. Appl Clin Inform. 2024;15(3):446-455. doi:10.1055/s-0044-1787007

8. Haque A, Milstein A, Fei-Fei L. Illuminating the dark spaces of healthcare with ambient intelligence. Nature. 2020;585(7824):193-202. doi:10.1038/s41586-020-2669-y

9. Tierney AA, Gayre G, Hoberman B, et al. Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catal Innov Care Deliv. 2024;5(3). doi:10.1056/cat.23.0404 ‌

10. Owens LM, Wilda JJ, Grifka R, Westendorp J, Fletcher JJ. Effect of ambient voice technology, natural language processing, and artificial intelligence on the patient-physician relationship. Appl Clin Inform. 2024;15(4):660-667. doi:10.1055/a-2337-4739

11. Zhang NJ, Seblega B, Wan T, Unruh L, Agiro A, Miao L. Health information technology adoption in U.S. acute care hospitals. J Med Syst. 2013;37(2):9907. doi:10.1007/s10916-012-9907-2

12. Nong P, Adler-Milstein J, Apathy NC, Holmgren AJ, Everson J. Current use and evaluation of artificial intelligence and predictive models in US hospitals. Health Aff (Millwood). 2025;44(1):90-98. doi:10.1377/hlthaff.2024.00842

13. Poon EG, Lemak CH, Rojas JC, Guptill J, Classen D. Adoption of artificial intelligence in healthcare: survey of health system priorities, successes, and challenges. J Am Med Inform Assoc. 2025;32(7):1093-1100. doi:10.1093/jamia/ocaf065

14. Bobek Tieché M. Most common hospital EHR systems by market share. Definitive Healthcare. January 10, 2024. Accessed April 28, 2025. https://web.archive.org/web/20250425182528/https://www.definitivehc.com/blog/most-common-inpatient-ehr-systems

15. AHA Annual Survey Database. American Hospital Association. Accessed July 1, 2025. https://www.ahadata.com/aha-annual-survey-database

16. Gupta SR, Wojtynek JE, Walton SM, et al. Monitoring of pharmacy staffing, workload, and productivity in community hospitals. Am J Health Syst Pharm. 2006;63(18):1728-1734. doi:10.2146/ajhp060012 ‌

17. Adler-Milstein J, Holmgren AJ, Kralovec P, Worzala C, Searcy T, Patel V. Electronic health record adoption in US hospitals: the emergence of a digital “advanced use” divide. J Am Med Inform Assoc. 2017;24(6):1142-1148. doi:10.1093/jamia/ocx080

18. Reinhardt UE. The economics of for-profit and not-for-profit hospitals. Health Aff (Millwood). 2000;19(6):178-186. doi:10.1377/hlthaff.19.6.178

19. Adler-Milstein J, Everson J, Lee SYD. EHR adoption and hospital performance: time-related effects. Health Serv Res. 2015;50(6):1751-1771. doi:10.1111/1475-6773.12406

20. Adler-Milstein J, Jha AK. HITECH Act drove large gains in hospital electronic health record adoption. Health Aff (Millwood). 2017;36(8):1416-1422. doi:10.1377/hlthaff.2016.1651