Highlights
Early implementation of an ambient artificial intelligence scribe in the emergency department was associated with shorter on-shift documentation time, lower total electronic health record time, and shorter notes.
Adoption was modest overall, occurring in 11.2% of eligible encounters, and was highly skewed: 35 of 92 attendings used the tool, with a small number of high-frequency users accounting for most ambient encounters.
Use clustered in telemedicine and vertical-care zones and among lower-acuity, noninterpreted encounters, suggesting that clinicians selectively deployed the tool in settings perceived to be operationally easier or lower risk.
The findings support workflow efficiency potential but also highlight important implementation barriers, equity concerns, and unanswered questions about note quality, generalizability, and downstream clinical impact.
Background
Clinical documentation is a major source of time burden in modern emergency care. Emergency physicians must synthesize fragmented histories, rapidly evolving diagnostic information, and risk-based medical decision-making under substantial throughput pressure. Although electronic health records have improved data accessibility and billing standardization, they have also expanded clerical demands, often shifting clinician time away from direct patient interaction. This is particularly relevant in the emergency department, where interruptions, multitasking, and compressed decision timelines make documentation friction especially costly.
Ambient AI scribes have emerged as a proposed response to this problem. Rather than requiring manual dictation or contemporaneous typing, these tools passively capture patient-clinician conversations and generate draft notes for physician review and editing. In theory, such systems may reduce keyboard time, improve eye contact, and lessen after-shift charting. However, real-world implementation in acute care environments remains poorly characterized. Emergency medicine presents a particularly stringent test case because encounters are brief, noisy, variable in acuity, and often involve interpreters, trainees, consultants, or parallel care processes.
Against this backdrop, Preiksaitis and colleagues examined real-world ambient AI scribe adoption and its association with documentation time in an academic emergency department. Their analysis is clinically valuable because it moves beyond proof-of-concept and evaluates actual use patterns through electronic health record audit logs rather than self-report alone.
Study Design and Methods
Design and setting
This was a retrospective observational study conducted at a tertiary academic medical center. The investigators evaluated adult emergency department encounters in which attending physicians had the option to use an ambient AI scribe to generate notes from patient-clinician conversations.
Population and eligibility
The study included adult emergency department visits managed by a single attending physician in core emergency department zones. Encounters involving human scribes were excluded to avoid confounding from a separate documentation support intervention. This design choice strengthened internal interpretability by focusing the comparison on ambient AI versus standard documentation workflows.
Exposure and comparator
The exposure of interest was ambient AI scribe use during the encounter. The comparator was standard documentation without ambient AI. Importantly, use was optional rather than randomized, making the study a pragmatic real-world implementation analysis rather than a causal effectiveness trial.
Outcomes
The main outcomes were derived from electronic health record audit logs and included documentation time during the shift, documentation time after the shift, total electronic health record time, and note length. Audit-log methodology offers an important strength because it provides behavior-based workflow data at scale, although it cannot fully capture cognitive workload, editing intensity, or note quality.
Analytic focus
The authors summarized adoption by physician, emergency department zone, and patient acuity, then compared median workflow measures between ambient and standard encounters. Because the abstract reports medians and differences rather than adjusted effect estimates, the results should be interpreted as associations in a selected population rather than proof that the AI tool itself caused the observed time savings.
Key Findings
Adoption was low overall and concentrated among a minority of physicians
Among 8,740 eligible encounters, 976 used ambient AI, corresponding to 11.2% of visits. Of 92 attending physicians, 35, or 38%, used the tool at least once. Adoption was therefore not only modest in aggregate but also incomplete at the clinician level. More notably, use was highly skewed, with a small group of high-frequency users accounting for most ambient encounters.
This pattern is important operationally. New digital tools are often described in terms of average uptake, but average uptake can conceal substantial concentration of use among early adopters. In this case, the data suggest that workflow benefits were not broadly distributed across the physician workforce during early implementation. A few engaged clinicians appeared to carry most of the use signal, which may reflect enthusiasm, familiarity, case mix, local champions, or individual differences in trust and editing tolerance.
Ambient AI was used selectively rather than uniformly
Ambient AI encounters clustered in telemedicine and vertical-care zones, the latter representing chair-based ambulatory emergency care settings. Use was also more common in lower-acuity patients and in encounters not requiring interpreters.
This selective deployment is clinically intuitive. Lower-acuity encounters are typically more linear, less interruption-prone, and may involve more predictable histories and fewer high-stakes time-sensitive interventions. Telemedicine and vertical-care settings may also provide cleaner audio capture, fewer overlapping conversations, and simpler conversational structure. By contrast, higher-acuity, resuscitation-oriented, interpreted, or procedurally complex encounters may be less amenable to ambient capture and may raise greater concerns about completeness, privacy, or transcription fidelity.
The implication is that the observed efficiency gains likely reflect not only the technology itself but also the contexts in which clinicians felt comfortable using it. That does not diminish the practical value of the findings, but it does limit generalizability across the full spectrum of emergency care.
Documentation time was shorter with ambient AI
The primary operational result was a reduction in on-shift documentation time. Median on-shift documentation time was 2 minutes 45 seconds for ambient encounters versus 3 minutes 50 seconds for standard encounters, an absolute difference of 1 minute 5 seconds and a relative reduction of 28%.
In a high-volume emergency department, this magnitude of per-patient savings can be meaningful. Even if the absolute difference seems modest at the encounter level, cumulative effects across a shift may translate into improved throughput, less work compression, or reduced cognitive switching between patient care and charting tasks.
Total EHR time was also lower
Median total electronic health record time was 8 minutes 39 seconds with ambient AI and 10 minutes 21 seconds with standard documentation, a relative reduction of 16%. This broader measure is particularly relevant because it captures more than note writing alone and may better reflect the overall digital workload associated with the encounter.
That said, total EHR time remains substantial in both groups, reminding readers that documentation is only one component of electronic work. Order entry, chart review, results management, messaging, and other EHR interactions are not eliminated by ambient note generation. Ambient AI may therefore be best understood as a partial workflow solution rather than a comprehensive remedy for digital burden.
Notes were shorter with ambient AI
Ambient-generated encounters were associated with shorter notes overall. Shorter documentation may be beneficial if it reflects reduced templating, less redundancy, and more concise synthesis. In emergency medicine, excessively long notes can obscure rather than clarify the clinician’s reasoning, especially when copied forward text crowds out the assessment and plan.
However, note length is an imperfect surrogate for note quality. Shorter notes could represent efficient summarization, but they could also risk omission of clinically or medicolegally important elements if physician review is insufficient or the system performs unevenly in complex cases. The abstract does not report direct quality assessment, coding outcomes, clinical safety endpoints, or physician editing burden, all of which would be essential for a fuller appraisal.
Clinical Interpretation
This study offers a pragmatic picture of how ambient AI may first enter emergency practice: not as a universally adopted platform, but as a selectively used tool that appears most attractive in lower-complexity environments. From an implementation perspective, that is a realistic and perhaps expected adoption curve. Clinicians often pilot new documentation technologies in cases where failure is less consequential and the workflow is easier to control.
The central signal is encouraging. When used, ambient AI was associated with less documentation time and lower total EHR time. For emergency departments struggling with clinician burnout, staffing pressure, and charting inefficiency, even incremental reductions in digital workload matter. If validated across settings, these gains could improve physician experience and potentially preserve more time for bedside communication.
At the same time, the adoption pattern suggests that the current generation of ambient AI may not yet be equally useful across emergency medicine’s full operational range. The underrepresentation of interpreted encounters is especially important. Language-concordant and interpreter-mediated visits are common in many emergency departments, and any tool that performs less well in these settings could inadvertently widen workflow inequities rather than reduce them. Similarly, lower uptake in higher-acuity cases means the technology’s value in precisely the most cognitively demanding encounters remains uncertain.
Strengths and Limitations
Strengths
The study’s major strengths include its real-world setting, use of electronic health record audit logs, and focus on actual clinician behavior rather than hypothetical acceptance. The sample size was substantial, with 8,740 eligible encounters, and the authors examined both adoption and workflow outcomes, which is critical for understanding implementation in practice.
Limitations
The main limitation is confounding by indication. Because physicians chose whether to use ambient AI, ambient encounters were likely systematically different from standard encounters. The clustering in low-acuity, noninterpreted, telemedicine, and vertical-care visits strongly suggests selection bias. Therefore, the shorter documentation time may partly reflect easier case mix rather than the independent effect of the tool.
The study was also conducted at a single tertiary academic center, which may limit external validity. Academic emergency departments often have distinct staffing models, documentation cultures, trainee involvement, and digital infrastructure. Additionally, the abstract does not report outcomes related to note accuracy, billing integrity, medico-legal sufficiency, patient satisfaction, physician burnout, or safety events. These omissions are understandable in an initial implementation study but are highly relevant for policy and procurement decisions.
Finally, audit logs quantify time in the EHR, not mental workload. A physician may spend less time typing yet more time reviewing and correcting AI-generated drafts. Without direct usability assessment, one cannot assume all time saved translates into less cognitive burden.
Implications for Practice and Health Systems
For clinicians, the study suggests that ambient AI scribes may already offer workflow benefit in selected emergency department scenarios, especially lower-acuity ambulatory care streams and telemedicine contexts. For department leaders, the findings underscore that rollout success should not be judged solely by whether the tool exists, but by who uses it, where it is used, and in which patient subgroups it appears feasible.
Implementation strategies may need to include targeted onboarding, specialty-specific prompt refinement, interpreter-compatible workflows, and transparent quality monitoring. Health systems should also resist assuming that early user enthusiasm equates to enterprise readiness. Concentrated adoption among a few physicians can create an overly optimistic impression unless performance is assessed across diverse users and encounter types.
Future studies should ideally include multicenter comparisons, adjusted analyses, randomized or crossover workflow trials, and direct review of note quality and safety. Questions of equity, bias, privacy, and reimbursement also deserve explicit study. In emergency medicine, where documentation serves clinical, operational, legal, and billing functions simultaneously, efficiency gains must be balanced against reliability.
Conclusion
Preiksaitis and colleagues provide an informative early look at ambient AI scribe use in emergency care. In this academic emergency department, adoption was limited overall and concentrated among a minority of physicians, with use favoring lower-acuity, noninterpreted encounters in telemedicine and vertical-care settings. When used, ambient AI was associated with shorter on-shift documentation time, lower total electronic health record time, and shorter notes.
The study supports ambient AI as a promising documentation support tool, but not yet a universally adopted solution. Its apparent benefits are real enough to warrant further evaluation, while its selective uptake highlights unresolved challenges in generalizability, complexity tolerance, and equitable deployment. For emergency medicine, the next phase of evidence should move beyond time savings alone to determine whether ambient AI can improve documentation quality, clinician experience, and patient care without introducing new risks.
Funding and Trial Registration
Funding information was not provided in the abstract. ClinicalTrials.gov registration was not reported, which is typical for retrospective observational workflow studies.
Citation
Preiksaitis C, Alvarez A, Winkel M, Karamatsu M, Brown I, Sama N, Morris L, Lee JY, Gubbels A, Wahl E, Frye A, Rose C. Ambient Artificial Intelligence Scribe Adoption and Documentation Time in the Emergency Department. Annals of Emergency Medicine. 2026-02-10;87(5):569-574. PMID: 41665590. Available at: https://pubmed.ncbi.nlm.nih.gov/41665590/