Response to “Ambient artificial intelligence scribes in the emergency department”

We thank Pelletier and colleagues for their careful commentary on our report of early ambient artificial intelligence (AI) scribe adoption in the emergency department (ED) (1,2). Their central interpretive claim is that our findings describe voluntary adoption in one implementation setting rather than establish causal effect. We are thankful for the chance to respond, first because we agree (and stated such in our paper), and second because the analysis they astutely recommend was already underway and has since been published (3).

Pelletier et al. propose that a stronger design would have used ambient and standard encounters from the same implementation period in a mixed-effects model adjusting for acuity, care zone, disposition, and interpreter need, with physician-level random effects. That is the analysis we conducted. In a 13-month cohort of 10,344 encounters managed by 100 attending physicians, over which adoption matured from 11.2% to 18.2%, ambient scribe use was associated with a 72.6-second adjusted reduction in on-shift documentation time per encounter [95% confidence interval (CI): 63.8 to 81.4; P<0.001] (3). What we would emphasize is not the estimate but its stability. The adjusted coefficient (−73 seconds) is nearly identical to the unadjusted median difference reported in Annals of Emergency Medicine (−65 seconds). Had case selection generated the original association, adjustment for the covariates Pelletier et al. name should have attenuated it substantially. However, it did not.

Two further analyses address their specific concern that early adopters differ in documentation style. First, a within-clinician placebo permutation test, in which the ambient indicator was shuffled 200 times across each physician’s own encounters, produced a null distribution centered at zero (mean −0.8 seconds); because each physician serves as his or her own comparator, stable individual differences in narrative preference cannot generate that contrast. Second, physicians using the tool above the cohort mean and those using it less realized comparable per-encounter savings (−71.6 vs. −64.2 seconds; interaction P=0.51), which is difficult to reconcile with an account in which the benefit belongs to a selected group of innovators rather than to the tool itself. Negative control outcomes on non-documentation electronic health record (EHR) activities were largely null. None of this establishes causation, and we continue to frame these findings as adjusted associations.

We would offer one correction of emphasis. Pelletier et al. conclude that our study is most representative of a fast track or urgent care environment. The zone-stratified results argue otherwise. Adjusted savings were largest in telemedicine (109.9 seconds) and were 69.4 seconds in main ED stretcher beds, whereas the smallest effect appeared in triage and lobby, the zone most closely resembling fast track (3). The gradient tracks workflow linearity and interruption frequency rather than acuity, and in the Annals of Emergency Medicine data the most pronounced median difference occurred at Emergency Severity Index level 2 (2). Ambient documentation appears to underperform where patient interactions are brief and task switching is high, which is a different limitation from the one they describe. Of course, many different effects are likely happening simultaneously, depending on the care environment, which we showed.

Nevertheless, several of their criticisms remain unanswered. The restriction to single-attending encounters persists in our newer work, and it continues to exclude the handoff-heavy, trainee-involved, and resuscitation encounters that often distinguish ED practice; whether these tools benefit residents is understudied though important. Their proposal for an interrupted time-series analysis of shift-level documentation time is likewise not addressed by our temporal adjustments, which operate at the encounter level. Whether making ambient documentation available reduces total clinician burden across a heterogeneous shift is a distinct question from whether it reduces time on the encounters where it is used, and it is the more relevant question for operational leaders. We think it is the better next study.

Their most important observation, however, concerns measurement. If a clinician spends 15 seconds deciding how best to characterize a presentation, Epic’s five-second inactivity threshold does not count it, and editing performed in vendor web portals falls outside audit capture altogether (1,4,5). We would add that our own secondary outcomes support rather than resolve this concern. After-shift documentation time increased by 9.1 seconds per encounter, and the reduction in total EHR time (56.6 seconds) was smaller than the reduction in note-specific time (72.6 seconds), a pattern more consistent with “displacement” of effort than with its “elimination” (3). Audit logs measure keystroke-adjacent activity; documentation burden is at least partly cognitive. Convergent designs pairing log-derived time with cognitive load and clinician-reported effort are needed before the field treats the two as interchangeable.

Finally, we agree that a one-minute per-encounter difference may be operationally uninteresting on its own. Our adjusted estimate corresponds to roughly 24 minutes across a 20-patient shift, which for a clinician working three to four shifts weekly approaches the 1.5 hours per week that Pelletier’s own work associates with meaningful reductions in burnout (6). We note that arithmetic cautiously, since we measured no wellbeing outcome and the threshold was derived in an ambulatory pediatric care setting. It does suggest the effect is at least in the range where wellbeing endpoints become worth measuring. We agree with Pelletier et al. without reservation on the call for randomized comparisons against both conventional documentation and human scribes, with note quality, coding accuracy, cost, and clinician and patient wellbeing as endpoints (7-9) to best understand the impact of this technology on the evolving practice of medicine.

Acknowledgments

None.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Journal of Medical Artificial Intelligence. The article did not undergo external peer review.

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-0141/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Pelletier JH, Wolski TP Jr, Rush SZ. Ambient artificial intelligence scribes in the emergency department. J Med Artif Intell 2026;9:67.
2. Preiksaitis C, Alvarez A, Winkel M, et al. Ambient Artificial Intelligence Scribe Adoption and Documentation Time in the Emergency Department. Ann Emerg Med 2026;87:569-74. [Crossref] [PubMed]
3. Preiksaitis C, Alvarez A, Winkel M, et al. Ambient AI Scribes and Emergency Department Documentation Burden: Retrospective Cohort Study. JMIR AI 2026;5:e92193. [Crossref] [PubMed]
4. Rule A, Melnick ER, Apathy NC. Using event logs to observe interactions with electronic health records: an updated scoping review shows increasing use of vendor-derived measures. J Am Med Inform Assoc 2022;30:144-54. Erratum in: J Am Med Inform Assoc 2023 Jan 18;30(2):408. [Crossref] [PubMed]
5. Guo Y, Hu D, Yang Z, et al. What do clinicians edit in ambient AI-drafted clinical documentation? A qualitative content analysis. J Am Med Inform Assoc 2026;33:1457-65. [Crossref] [PubMed]
6. Pelletier JH, Watson K, Michel J, et al. Effect of a generative artificial intelligence digital scribe on pediatric provider documentation time, cognitive burden, and burnout. JAMIA Open 2025;8:ooaf068. [Crossref] [PubMed]
7. Morey J, Jones D, Walker L, et al. Ambient Artificial Intelligence Versus Human Scribes in the Emergency Department. Ann Emerg Med 2026;87:561-8. [Crossref] [PubMed]
8. Dutta S, Guan-Ting You J, Dunham L, et al. Medical Scribe and Ambient Artificial Intelligence Impact on Emergency Physician Documentation Burden and Clinical Productivity. Ann Emerg Med 2026;S0196-0644(26)00266-0. [Epub ahead of print]. [Crossref] [PubMed]
9. Gottlieb M, Palter J, Westrick J, et al. Effect of Medical Scribes on Throughput, Revenue, and Patient and Provider Satisfaction: A Systematic Review and Meta-analysis. Ann Emerg Med 2021;77:180-9. [Crossref] [PubMed]