Blog

Faster Notes, Same Workload? What the Beth Israel Lahey Heidi Study Means for Psychiatry

Published 2026-11-02 · Last updated 2026-11-02

Changelog
  • 2026-11-02: Published.

A large health system has now measured what an ambient AI scribe does to clinicians' days, and the result is more useful than any vendor case study: notes got faster, notes got longer, and workloads did not get lighter. The study, a quality improvement analysis published online in *Annals of Internal Medicine* in late October 2026, followed outpatient clinicians at Beth Israel Lahey Health, a 14-hospital system, after they enrolled in Heidi's ambient scribe [1]. For clinicians weighing a scribe purchase, it is the best check yet on the claim that the tool gives you your evenings back, because it measures time in the chart, note closure, and productivity.

What the study actually measured

Researchers analyzed more than 1.6 million outpatient encounters (1,637,227) involving 902 clinicians who enrolled in Heidi after it became available system-wide in January 2025, using note data from July 2024 through January 2026 [1][2]. Enrollment was voluntary, and Heidi ran as a standalone platform, not integrated with the EHR [2]. The main outcome was note-active time, meaning minutes with a note open in the EHR, which is narrower than "time spent documenting." Time spent editing a draft inside Heidi before it reached the EHR was not captured, and note accuracy was not assessed [2].

Two other limits deserve equal weight. First, the design is an interrupted time series around each clinician's enrollment month; it shows an association, not proof that Heidi alone caused the changes, and the authors flag possible unmeasured confounding and their inability to tell AI-scribed notes from clinician-written ones after enrollment [1]. Second, one co-author is affiliated with Heidi Health, and the paper reports no primary funding source [4]. Neither caveat invalidates the findings, but they cap what the study supports: a real-world signal, not a controlled trial.

What improved, and what did not

Enrollment was associated with an immediate drop of about 1.2 minutes of note-active time per encounter [1]. In raw terms, median note-active time fell from 21 to 18.6 minutes per encounter, while median note length rose from 4,854 to 5,187 characters [3]. But the savings stopped at the note itself: the study found no differences in same-day note closure, after-hours note closure, or work relative value units [1].

The most quoted figure needs its context attached. High users, those who used Heidi in 55% of encounters during their first three months, cut note-active time by about 2.5 minutes per encounter, which the authors estimate at roughly 25 minutes for a clinician seeing 10 patients per session [1][2]. Low users, who used it in 3% of encounters, and the 297 enrolled clinicians who never used it saw minimal change [2]. High users' notes also grew longer than those of nonusers and low users [1]. The clinicians who got the most out of the tool were also the ones whose notes expanded the most.

Why the gains were uneven

Three features of the deployment explain the pattern.

  • Use it or lose it. Benefit tracked usage, and a third of enrollees (297 of 902) never used the tool at all [1]. A scribe license that sits unused is not a saving, and a trial nobody on the team adopts tells you nothing about the tool.
  • Standalone means a transfer step. Because Heidi ran outside the EHR, every note still had to travel from the scribe app into the chart [2]. That transfer, and the editing inside Heidi that the study could not see, is exactly where saved minutes can leak away. Our comparison of standalone versus EHR-bundled AI notes works through the same trade-off: a standalone tool keeps your EHR choice independent, but it adds a handoff a bundled tool does not.
  • Longer is not better by default. Median notes grew by roughly 330 characters [3]. In psychiatry, where the record is mostly reported speech, attributed history, and clinical inference, extra length can mean extra review burden, or extra carried-forward language, rather than extra quality. The data fit a story in which scribes partly relocate documentation work rather than remove it, better than the simpler story in which faster drafting equals a lighter day.

What this does not prove for psychiatry

The study population was general outpatient clinicians, not psychiatrists or therapists, so applying its numbers to a 50-minute therapy session or a medication-management visit is an extrapolation [1]. Psychiatry notes also carry risks the study did not measure: a differential charted as a diagnosis, collateral merged into the patient's history, or an invented mental status finding. Our clinical pilot scored Heidi's notes on synthetic psychiatric encounters at 96.0 with zero critical flags, but with one scored run per scenario, synthetic audio, and no second independent psychiatrist review, it is a signal under test conditions, not a guarantee for a real caseload. That distinction is also why Heidi's position on our site needs two numbers, not one. On our research scorecard, Heidi (88/100) is the top-scored vendor, on documentation measures, not clinical performance. Our Heidi review carries both figures: the 88 research score and the separate 96.0 pilot score. The study adds a third kind of evidence, real-world time effects, and it is the least flattering of the three. A buyer should hold all three at once.

How to read the next time-savings claim

The study helps separate two kinds of vendor time claims. Heidi's own psychiatry FAQ, an AI-generated answer on its site, describes psychiatrists reclaiming evenings and cites one physician's account of cutting daily charting time by roughly 50 to 60% [5]. The Beth Israel Lahey data found no change in after-hours closure [1]. Only the study figure comes from 902 clinicians and 1.6 million encounters, and only it measured after-hours work directly. Treat a single-clinician account as a reason to trial, not as a saving to budget.

A trial that measures what matters

If you trial a scribe, measure the outcomes vendors rarely headline:

  • Editing time, not just drafting time. Log the minutes you spend reviewing and correcting each draft inside the scribe app before it reaches the chart. The study's blind spot is the likeliest place for your savings to disappear.
  • After-hours closure, not just same-day speed. Track what share of notes close after hours in the trial weeks versus the weeks before. If notes are drafted faster but signed later, the burden moved; it did not shrink.
  • Note length against note quality. Compare the draft's length with what you would have written, and check the additions against the session. Our pre-sign checklist targets the psychiatry-specific failure modes (differentials upgraded to diagnoses, collateral folded into the patient's history, invented findings) that longer notes can hide.
  • Usage across the team, not just the enthusiast. The study's nonusers are a warning: report adoption per clinician, not per practice.
  • The handoff cost. If you run standalone, time the transfer step into your EHR. If your EHR's bundled tool is the alternative, weigh that handoff against the switching cost of coupling the two decisions, priced out in our psychiatric EHR comparison across 13 EHRs.

The bottom line: This peer-reviewed study says an ambient scribe can cut minutes per note, especially for clinicians who actually use it, and still leave after-hours work and productivity unchanged. Buy the tool for the minutes it demonstrably saves you, measured your way, and keep the signature review as the step where the note becomes yours.

Sources

  1. Annals of Internal Medicine, "Impact of Unrestricted Artificial Intelligence Scribe Deployment in a Single Health System on Clinician Workload and Productivity: A Quality Improvement Study" (published online October 2026), https://doi.org/10.7326/ANNALS-25-05294 (abstract opened 2026-11-02)
  2. Becker's Hospital Review, "Heavy AI scribe users at Beth Israel Lahey cut note time 2.5 minutes," https://www.beckershospitalreview.com/healthcare-information-technology/ai/heavy-ai-scribe-users-at-beth-israel-lahey-cut-note-time-25-minutes/ (opened 2026-11-02)
  3. MedPage Today, "AI Scribes Improved Note-Taking Time but Not Productivity, Study Finds" (October 26, 2026), via RamaOnHealthcare, https://ramaonhealthcare.com/ai-scribes-improved-note-taking-time-but-not-productivity-study-finds/ (opened 2026-11-02)
  4. What AI Found, "System-wide AI scribe deployment and clinician documentation time at a 14-hospital health system," https://whataifound.org/finding/2026-10-27-heidi-ai-scribe-health-system (opened 2026-11-02)
  5. Heidi Health, "How can a psychiatrist use Heidi Scribe to save time?" (vendor FAQ, AI-generated answer), https://www.heidihealth.com/search/faq/how-can-a-psychiatrist-use-heidi-scribe-to-save-time (opened 2026-11-02)

Who curates this site

Curated by a practicing, board-certified psychiatrist, not named to avoid conflicts. From the site data: 19 vendors tracked, 354 distinct external source pages linked. Last site update 2026-11-20; last targeted vendor verification 2026-11-12; last complete research audit 2026-11-04. How we work · Editorial policy.