AI and Physician Burnout: Honest Math on a Real Problem

The fourth-year resident I rounded with last Tuesday told me she had not finished her notes from Sunday's shift. It was Tuesday afternoon. The notes were three days old. She had been on shift for thirty-six of the intervening seventy-two hours. She is one of the best clinicians I have trained in a decade. She is also, by every published instrument we have, burned out. And the most honest conversation I can have with her right now is the one about what AI actually does for that problem and what it does not.

There is finally good data on AI and physician burnout. The 2025-2026 literature is large enough, multi-site enough, and methodologically clean enough that we can stop trading anecdotes and start doing the math. The math is genuinely encouraging on one axis and deeply uncomfortable on another. Emergency physicians, residency program directors, and ED medical directors need to understand both, because the wrong read on this evidence will spend a lot of money on a tool that solves the wrong half of the problem.

After eighteen months of running ambient AI scribe pilots, ambient documentation tools, and AI-augmented workflows in emergency department settings, I want to walk through what the evidence actually says, what it does not say, and what the operational and leadership response should be for ED medical directors evaluating these tools right now.

What the 2025-2026 Evidence Actually Shows

The headline study is the multi-site JAMA paper Rotenstein and colleagues published in April 2026. Across five academic medical centers, ambient AI scribe adoption was associated with a thirteen-minute reduction in daily total EHR time and a sixteen-minute reduction in daily documentation time. That is a three percent and ten percent relative reduction respectively. Clinicians who used the scribe for more than half of patient encounters experienced roughly double the EHR time reduction and triple the documentation time reduction of average users. Productivity rose by about half a visit per week.

The well-being signal is more impressive than the time signal. The 2025 JAMA Network Open survey study across Mass General Brigham and Emory Healthcare reported a 21.2% absolute reduction in burnout prevalence at MGB after eighty-four days of ambient documentation use, and a 30.7% absolute increase in documentation-related well-being at Emory. A 2025 study Shah and colleagues published in JAMIA showed a statistically significant drop on the Stanford Professional Fulfillment Index work-exhaustion subscale among DAX Copilot users. A 2026 JMIR Medical Informatics study reported a fifteen percent reduction in per-consultation documentation time and a 10.6% increase in measured eye contact during patient encounters when the scribe was active.

That body of evidence is consistent enough that I am no longer skeptical of the underlying claim. AI scribes do reduce documentation burden and they do correlate with measurable reductions in burnout and improvements in well-being scores. The mechanism is the cognitive offload. Stults and colleagues reported NASA Task Load Index mental-demand scores dropping from 12.2 to 6.3 and overall task effort scores dropping from 12.5 to 7.4 with ambient scribe use. Physicians stopped carrying the documentation in their heads between patients. That is the win.

The Uncomfortable Half of the Math

But the same studies that report those wins also report the ceiling, and the ceiling is the part the vendor presentations skip.

The thirteen to sixteen minutes a day the JAMA multi-site study found is real, but it is not the four hours of pajama time the average emergency physician is actually carrying. The relative reduction is three to ten percent of total EHR time, not fifty. The 21.2% absolute burnout reduction at MGB was reported in a survey-study population with a low response rate that, by the authors' own caveat, may have over-represented enthusiastic adopters. Only thirty-two percent of clinicians in the JAMA multi-site cohort used the scribe in more than half their encounters, which means most users were not getting the dose that produced the biggest effect. The 2026 cardiovascular review from PMC noted that AI scribes consistently reduce documentation burden but also produce documentation omissions and occasional clinically significant hallucinations that have to be caught on review, which is itself a cognitive load.

The deeper limit is that documentation is one driver of burnout, not the driver. The 2025-2026 burnout literature is converging on a model in which administrative load is roughly a third of the equation, with cognitive overload from interruptions and information density a second third, and what Talbot and Dean called moral injury — the structural mismatch between what physicians are asked to do and what they were trained to do — as the third and largest piece. The clinicians I have lost from emergency medicine in the last three years did not leave because of pajama time. They left because the system asked them to board admitted patients in hallway beds for fourteen hours, hold telephone consultants accountable with no leverage, and shoulder downstream responsibility for decisions they did not make. AI scribes do not touch that. For the leadership framing on this distinction — burnout as the symptom, moral injury as the wound — I would point you to my Medium essay on the topic: Moral Injury Is the Leadership Problem Emergency Medicine Keeps Misdiagnosing.

The other uncomfortable point is the productivity creep. The same JAMA multi-site study that reported burnout improvements also reported a 0.5-visit-per-week productivity increase. That number is small enough that no one is alarmed yet, but the operational logic of every health system that buys an AI scribe is the same: if the scribe gives clinicians back time, the system will eventually expect that time back in the form of throughput. The scribe vendor sells reduced burden; the chief financial officer reads the same study and sees recoverable RVUs. Without a hard contractual commitment from the institution that ambient scribe time savings will not be converted to additional patient load, the burnout gain is on a timer.

An Honest Operational Framework

Here is what I would put in writing before any ED commits to an enterprise ambient AI scribe deployment.

The scribe is a documentation tool, not a wellness program. Frame it that way internally. The institutional wellness committee should not own the rollout, because that framing sets the wrong expectation about what the tool can deliver. The chief medical informatics officer and the ED medical director should own it jointly, and the goal should be expressed in operational terms — documentation time per shift, after-hours charting, EHR time per encounter — not in burnout-score deltas.

Adoption is the dependent variable, not the independent one. The JAMA data is clear that the clinicians who use the scribe heavily get the biggest gain, and the clinicians who use it lightly get almost nothing. That means the rollout has to invest in the training, the workflow integration, and the post-go-live support that gets users from occasional use to majority-of-encounters use. Buying the license and announcing it at a department meeting does not produce the published outcomes.

Note review remains the physician's responsibility and the physician's risk. Every published study notes that AI scribes generate omissions and occasional hallucinations that require clinician review. The note that the physician signs is the physician's note, full stop. Your departmental policy needs to say that explicitly, and your training needs to cover the specific patterns of omission and hallucination the chosen vendor produces. Vendors will resist this framing. They are wrong to.

Time savings are protected, not converted. If the institution is unwilling to commit in writing that the time savings from the scribe will be returned to the clinician rather than absorbed by additional patient load, the deployment will erode burnout gains within twelve months. That commitment is a leadership decision, not a technology decision, and the ED medical director needs to make it explicit before the rollout, not after.

What You Should Be Doing Now

If your department is in active AI scribe procurement or pilot, audit the contract language around productivity expectations. The clinical wins in the published data depend on the time savings being protected, and that protection has to be in writing. Pull your last twelve months of after-hours documentation data and use it as the operational baseline, not burnout scores, so that the rollout can be evaluated on the variable the tool actually moves. Identify your power users in the pilot phase and ask them what their workflow looks like — those are the patterns that need to be standardized across the department, because the JAMA data shows that high-use clinicians are getting the published outcomes and average users are not. And recognize, explicitly, that the AI scribe is one tool addressing one driver of burnout. The boarding problem, the consultant-leverage problem, and the moral injury problem still need their own interventions, and none of them are software.

Dr. Chet's Take

I have stood in operations centers for air medical programs, telemedicine networks, and military medical task forces, and the pattern in all three settings is the same. Technology shows up promising to fix a people problem. It delivers a partial fix to a piece of the people problem. Leadership then has to decide whether the partial fix will be allowed to do its job or whether the system will absorb the savings and rebuild the same pressure on top of it. In every command I have served in, the difference between the technology delivering on its promise and the technology becoming another grievance was leadership discipline at the moment the savings showed up.

The published evidence on AI scribes is real and it is honest. The tool works, modestly, on the documentation half of the burnout equation. The wound underneath is leadership work — boarding, throughput, scope creep, scope mismatch, the structural conditions that drive moral injury. Buy the scribe if it earns the case. Do not let the institution sell it to your physicians as the answer to a problem the institution itself created.

AI Won't Wait. Neither Should You.

If your department is rolling out an ambient AI scribe and the conversation is being framed as a wellness initiative rather than a documentation tool with operational guardrails, you are exposed. Consider enrolling in my course: AI in Emergency Medicine: Becoming AI Bulletproof. The course walks through the procurement checklist for ambient documentation vendors, the contract language we use to protect clinician time savings from productivity creep, the departmental policy template for AI-generated note review, and the operational dashboard I run with our quality committee to evaluate scribe deployment on the variables that actually predict the published outcomes.

Learn more: AI in Emergency Medicine: Becoming AI Bulletproof

If you're an emergency physician (or any clinician treating patients daily) trying to understand how AI will actually impact your clinical practice — not just the hype — I put together a free practical guide. You can download it here: AI in EM Survival Guide.

Dr. Chester "Chet" Shermer, MD, FACEP is a Professor of Emergency Medicine, TeleHealth, HEMS and Critical Care Transport, and State Surgeon for the Army National Guard. He is the founder of Global MedOps Command and the creator of AI in Emergency Medicine: Becoming AI Bulletproof.

Also on Medium: Read more from Dr. Shermer on Medium.

Sources

1. JAMA (Rotenstein L, et al.), "Changes in Clinician Time Expenditure and Visit Quantity With Adoption of Artificial Intelligence-Powered Scribes: A Multisite Study," April 2026, https://www.massgeneralbrigham.org/en/about/newsroom/press-releases/ai-scribes-linked-to-modest-reductions-in-ehr-documentation-time
2. JAMA Network Open, "Ambient Documentation Technology in Clinician Experience of Documentation Burden," August 2025, https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2837847
3. Nature npj Digital Medicine, "Barriers and opportunities of scaling ambient AI scribes for clinical workflows," March 2026, https://www.nature.com/articles/s41746-026-02554-0
4. JMIR Medical Informatics, "Impact of an Ambient AI Scribe Among Clinicians and Patients," March 2026, https://medinform.jmir.org/2026/1/e85580
5. PMC / Cardiovascular Diagnosis and Therapy, "Transforming clinical documentation with ambient artificial intelligence scribes," January 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12973079/