This analysis synthesizes 7 sources published the week ending Oct 27, 2026. Editorial analysis by the PhysEmp Editorial Team.
Health system leaders bought ambient AI scribes expecting a productivity return. This week, two large health systems published peer-reviewed measurements of what the tools actually changed. The answer is minutes of documentation time and, in the one study that measured it, no change in work RVUs. That gap between the business case and the data is now the most practical question in AI in Physician Employment & Clinical Practice, because productivity targets, panel sizes and compensation formulas are being set against the business case.
The physicians most exposed are employed ones, which is to say most of them. The Physicians Foundation’s 2026 survey, as reported by Medical Economics, puts the share of physicians who are now employees of some kind at about 80%.
Two studies, one flat productivity line
The larger study comes from Beth Israel Lahey Health and was published Oct. 27 in Annals of Internal Medicine. Researchers tracked more than 1.6 million outpatient encounters from July 2024 through January 2026 across 902 clinicians enrolled in Heidi’s ambient scribe. According to the journal’s release, enrollment was associated with an immediate 1.2-minute drop in note-active time per encounter. High users saw about 2.5 minutes. Clinicians who used it rarely or never, including 297 who enrolled and never used it, saw almost nothing.
The productivity results are the part executives will read twice. Becker’s reports no significant change in after-hours note closures or in work RVUs. High users’ notes got longer, and they closed notes after hours more often than enrolled nonusers. The study did not capture time spent editing inside Heidi, and it did not assess note accuracy. One co-author is affiliated with Heidi, though the paper says the company had no role in the analysis.
Jefferson Health’s results, published Oct. 23 in NEJM Catalyst, look stronger on documentation and say little about output. Among 637 clinicians who crossed a usage threshold with Abridge, time to chart closure fell 25.3% and 39.3% in the system’s two EHR cohorts, and after-hours documentation fell 8.9% and 16.6%. Weekly progress note volume held steady or rose, but the study reported no visit-volume or financial results. Only about 34% of the 5,234 clinicians with access actively recorded visits.
The adoption numbers may matter more than the time savings. If roughly a third of clinicians with access use the tool, any productivity expectation applied across a whole department is being set by a minority of heavy users.
Where the saved minutes went
One detail from Jefferson deserves attention from anyone building a staffing model. Advanced practice providers regained 9.9 minutes of daily documentation time. Physicians regained 3.8. The study does not explain the gap. For an employer weighing whether to hire another physician or another APP, it is a number that will get noticed.
Kirsten Ostherr of Rice University, in an editorial accompanying the Annals paper, asked whether ambient scribes “relocate documentation burden rather than reduce it.” Gary Price, president of the Physicians Foundation, offered a plainer version to Medical Economics: the time scribes freed was absorbed by a backlog of other administrative work, which he calls the “stupid stuff.” He also said institutional and financial leaders were disappointed that the saved time did not produce the productivity they expected.
That disappointment is the employment risk. A second Medical Economics piece lays out what happens when AI’s value is measured in productivity: extra appointments, shorter visits and larger panels, with physicians who “aren’t always the ones who decide” where reclaimed time goes. The new data suggests that pressure would rest on a thin base. At BILH’s high-user rate, the authors estimate about 25 minutes saved across a 10-patient session. Work RVUs did not move, and that estimate comes before any time spent reviewing the draft.
In reviewing our own coverage recently, we may have given physician burnout less attention than it has earned, and the Physicians Foundation figures here suggest it deserves a closer look. The same survey found 49% of physicians said well-being concerns led them to consider or take steps such as cutting clinical hours or retiring early in the past year.
The review time nobody schedules
Neither study measured how long physicians spend checking what the scribe wrote, and that is where the liability sits. Medical Economics’ review of the legal exposure notes that an e-signature functions as an attestation that the physician reviewed the note. One evaluation it cites found omissions in 18% of physician-reviewed notes and hallucinations in 11.5%, even though 94.7% of sampled notes had no significant errors. UCSF’s Robert Wachter questioned whether he would stay attentive reviewing note No. 50 after 49 perfect ones. The same article reports that many vendor contracts cap the company’s liability at 12 months of fees.
A schedule that converts scribe savings into added visits removes the margin physicians would use to read the notes they sign. The time and the liability end up on opposite sides of the ledger.
For candidates evaluating offers, that turns into specific questions: whether productivity targets were raised when the scribe was introduced, whether the schedule assumes time for note review, and who carries the risk when an AI-drafted note is wrong. Recruiters will hear these more often as the studies circulate.
The metric is shifting from notes to whole encounters
Health system technology leaders are already moving past the note. Becker’s reports that CIOs now want AI that handles orders, referrals and prior authorization, the work that follows a visit. Tom Barnett of Baptist Memorial Health Care argues the right measure is how long the whole encounter takes to close, not physician documentation time alone. North Country Healthcare’s Darrell Bodnar says most tools write notes well but struggle with orders, referrals and charge capture.
That framing may help physicians. If the unit of value becomes a closed encounter, the minutes saved on a note matter less than the downstream tasks still waiting, which the BILH and Jefferson data suggest are where the time went. It also gives employers a new reason to raise targets once the next generation of tools arrives.
The number to watch is work RVUs per clinical hour at systems that publish scribe outcomes over the next year. If it stays flat while documentation time keeps falling, the argument for tying AI rollouts to higher productivity targets loses its evidence. If it rises only for APPs, staffing models will change before physician contracts do.
Sources
Impact of Unrestricted Artificial Intelligence Scribe Deployment in a Single Health System on Clinician Workload and Productivity (news release) – EurekAlert!
Heavy AI scribe users at Beth Israel Lahey cut note time 2.5 minutes – Becker’s Hospital Review
Ambient AI cut chart closure time up to 39% at Jefferson Health – Becker’s Hospital Review
Getting rid of the ‘stupid stuff’ — why AI scribes alone can’t fix physician burnout – Medical Economics
Will AI solve burnout — or just create a new source of it? – Medical Economics
Your AI scribe is saving you time. Is it also building the case against you? – Medical Economics
The EHR tasks CIOs want AI to replace – Becker’s Hospital Review