Weekly digest

Digest Issue #5

The Big Finding, Product Watch, Clinical Reality Check, Head-to-Head, and One Action This Week.

Last updated 2026-11-04

Published 2026-11-04. Sources opened 2026-11-03. Every claim sourced; see the methodology.

One short, sourced email a week: what changed in AI scribe pricing, privacy terms, and features.

The Big Finding

Faster notes, same workload: the first large peer-reviewed measurement of an AI scribe in routine practice. A quality-improvement study published in Annals of Internal Medicine on October 26, 2026, followed 902 outpatient clinicians at Beth Israel Lahey Health, a 14-hospital system, after they enrolled in Heidi's ambient scribe [1][2]. Across all enrolled clinicians, median note-active time (minutes with a note open in the EHR) fell from 21 to 18.6 minutes per encounter, while median note length rose from 4,854 to 5,187 characters [2]. But the savings stopped at the note itself: the study found no significant change in same-day note closure, after-hours note closure, or work relative value units [1].

The headline figure needs its context attached. High users, who used Heidi in 55% of encounters during their first three months, cut note-active time by about 2.5 minutes per encounter, roughly 25 minutes for a clinician seeing 10 patients per session [1]. Low users (3% of encounters) saw a 0.78-minute reduction, and the 297 enrolled clinicians who never used the tool after enrolling saw minimal change [2]. Caveats, stated plainly: the design is an interrupted time series around each clinician's enrollment, so it shows an association, not proof that Heidi alone caused the changes; time spent editing a draft inside Heidi before it reached the EHR was not captured, and note accuracy was not assessed; and one co-author is affiliated with Heidi Health, though the paper states the company had no role in data analysis or the decision to submit [1][2]. Heidi ran as a standalone platform, not integrated with the EHR, so every note still had to travel from the scribe app into the chart [2].

Why this leads the issue: Heidi is the vendor that leads our site on two separate measures, and this study adds a third, different kind of evidence. Heidi (88/100) is ranked first on our research scorecard on public-documentation measures (not clinical performance), and our clinical pilot scored Heidi's notes on synthetic psychiatric encounters at 96.0 with zero critical flags, under test conditions with one scored run per scenario, synthetic audio, and no second independent psychiatrist review [3]. The Beth Israel Lahey data measure real-world time effects, and they say: buy the tool for the minutes it demonstrably saves you, measured your way. Our full post works through the trial measures that matter: clinicalaiweekly.com/blog/heidi-beth-israel-lahey-study-psychiatry

Product Watch

  • Oncologists liked their scribe even though note closure barely moved. A prospective survey at Memorial Sloan Kettering Cancer Center, presented at the ASCO Quality Care Symposium on October 16 to 17, 2026, found clinicians reporting high or excessive note writing fell from 78% to 60% after the intervention, and 73% of users said they would recommend the scribe to others, despite only small changes in after-hours work and note closure [4]. Caveats: this is a conference abstract, not a peer-reviewed paper; uptake was lower than expected (only 19.8% of respondents used the scribe); and the center launched a new EHR at the same time, which the authors say could account for the dramatic after-hours drop among never-users [4]. The pattern rhymes with the Beth Israel Lahey result: perceived benefit and measured workload relief are not the same thing.
  • No new scorecard vendor this week. The scorecard still covers 19 tracked vendors, 17 ranked, with scores as of 2026-10-23 [5]. One pricing note from the live scorecard: Twofold Health's Personal plan is currently listed at $21/month billed annually as a promotion, against a standard $69/month monthly price [5]. (Update Nov 4: Twofold has since ended the $21 promo; current pricing starts at $49/mo billed annually.) Scores and ranks are unchanged.

Clinical Reality Check

Psychiatry was near the bottom of scribe adoption in a systemwide rollout, and that is information, not a verdict on the tool. A second study published October 26, 2026, this one in npj Digital Medicine, tracked 2,901 outpatient clinicians at one integrated academic health system (Johns Hopkins) from December 2024 to May 2025 [6]. Adoption varied sharply by specialty: community general pediatrics led at 69% of clinicians using the scribe at least once, while psychiatry sat at 19%, alongside academic general surgery (16%) and obstetrics and gynecology (18%) [6]. Where the tool fit, it became routine fast: in some groups, use reached roughly 80% to 90% of eligible visits by May 2025. But some clinicians used it early and then stopped, so counting everyone who ever tried it hides how many quietly gave it up [6]. Caveats: one health system, six months, usage records only. The study measured how much the tool was used, not whether it saved time or improved notes, and it cannot prove why some specialties used it more than others [6]. For behavioral health, the question it raises is practical: is psychiatry's low adoption a format problem, a workflow problem, or a trust problem? A trial that reports adoption per clinician, not per practice, is how you find out.

A scribe can only chart what was spoken, and in psychiatry, much of the exam is not spoken. A Perspective published in the Journal of General Internal Medicine on October 29, 2026, argues that audio-based documentation has a structural blind spot: it captures what is said, not what clinicians observe without verbalizing or judge without articulating [7]. For psychiatry, the implication is direct. Affect, psychomotor activity, eye contact, and the clinician's rapid, experience-based judgment about risk or rapport often never enter the spoken record, and a note built from speech alone may systematically underrepresent them [7]. The author proposes evaluating scribes for reasoning fidelity, not transcription accuracy alone, and teaching learners the distinction between tacit judgment and bias [7]. This is a Perspective, an argument rather than a measured result, but it names the check a signing clinician should run on every draft: what did you observe that the transcript never heard?

Head-to-Head

Standalone AI scribe or EHR-bundled AI notes: how to choose. This is the question our October 29 post works through in full, and the Beth Israel Lahey study sharpens it: Heidi ran standalone there, and the transfer step into the EHR is exactly where saved minutes can leak [2]. Choose bundled when your EHR is a keeper and its AI tool fits your note formats and BAA questions: the note starts and ends in the chart, with no second app and no transfer step, but changing EHRs later means changing your note tool at the same time. Choose standalone when format depth, a separate BAA with the scribe vendor, or freedom to change EHRs matters more: a second invoice, but one that survives an EHR change [8]. The comparison that matters is documented monthly costs and add-ons for one prescriber: in our psychiatric EHR comparison (13 EHRs), estimated published monthly costs run from about $104 for TheraNest (74, first on our EHR research score) and $137 for OptiMantra to about $221 for CharmHealth, before any standalone scribe is added [8][10]. One bundled example carries its caveat with it: Tebra's AI Note Assist supports Therapist Initial, Therapist Progress, and Psych Initial formats, and its headline time figures (documentation time from about 23 to about 14 minutes per note) are vendor-reported from a customer survey of 2,297 paired responses, self-reported rather than directly measured [8]. Read the full post: clinicalaiweekly.com/blog/standalone-vs-ehr-bundled-ai-scribe

One Action This Week

Before you sign your next AI-drafted note, add back what the microphone never heard. Read the draft against your own mental status exam: affect, psychomotor activity, eye contact, rapport, and any risk judgment you formed but did not say aloud. If the draft is silent on those, write them in yourself before signing. A scribe that only heard the conversation cannot supply them [7]. Then run the trial measures from this issue's Big Finding: log your editing time inside the scribe app, track after-hours closure against the weeks before the trial, and report adoption per clinician, not per practice [1]. If notes are drafted faster but signed later, the burden moved; it did not shrink.

Also new on the site

  • New blog post, November 2: "Faster Notes, Same Workload? What the Beth Israel Lahey Heidi Study Means for Psychiatry": the full version of this issue's Big Finding, including the three-number Heidi framing (88 research score, 96.0 pilot score, real-world time effects) and a trial checklist [3].
  • New blog post, October 29: "Standalone AI Scribe or EHR-Bundled AI Notes: How to Choose": formats, BAA questions, cost stacking, and switching costs, the full version of this issue's Head-to-Head [8].
  • Corrections, October 29 to 30, stated plainly: the Best Picks counts and review date were refreshed (picks unchanged) with the clinical pilot shown as a separate section, and the EHR comparison, estimated cost table, and cost calculator now mark 8 EHRs whose published pricing is incomplete [9]. And on October 30, the pilot methodology page was corrected: it had said that a second clinician reviewed every critical error before any vendor was named. That was not accurate: each note is scored as a draft against the locked checklist, then adjudicated by a single psychiatrist scorer, and critical errors are published as draft-adjudicated with no second independent reviewer yet [9]. Pilot scores, critical flags, research scores, and ranks are unchanged.
  • Scorecard unchanged: 19 tracked vendors, 17 ranked, scores as of 2026-10-23 [5]. Research scores rate public vendor documentation, not note accuracy or clinical outcomes.

The bottom line: The best evidence this week says scribes save minutes per note for clinicians who actually use them, leave after-hours work largely unchanged, and cannot chart what was never spoken, so measure the trial yourself, and sign only the note you have completed.

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 26, 2026), https://doi.org/10.7326/ANNALS-25-05294 (opened 2026-11-03)
  2. Becker's Hospital Review, "Heavy AI scribe users at Beth Israel Lahey cut note time 2.5 minutes" (October 26, 2026), https://www.beckershospitalreview.com/healthcare-information-technology/ai/heavy-ai-scribe-users-at-beth-israel-lahey-cut-note-time-2-5-minutes/ (opened 2026-11-03)
  3. ClinicalAIWeekly, "Faster Notes, Same Workload? What the Beth Israel Lahey Heidi Study Means for Psychiatry" (blog, published 2026-11-02), https://www.clinicalaiweekly.com/blog/heidi-beth-israel-lahey-study-psychiatry (opened 2026-11-03)
  4. Healio, "Oncologists give ambient AI scribes positive reviews" (October 27, 2026; Baldwin-Medsker A, et al. Abstract 673, ASCO Quality Care Symposium, October 16 to 17, 2026), https://www.healio.com/news/hematology-oncology/20261027/an-exciting-time-oncologists-give-ambient-ai-scribes-positive-reviews (opened 2026-11-03)
  5. ClinicalAIWeekly, "AI scribe research scorecard" (19 tracked, 17 ranked; scores as of 2026-10-23), https://www.clinicalaiweekly.com/scorecard (opened 2026-11-03)
  6. Azad TD, et al., "Systemwide adoption patterns of ambient AI scribe technology in an integrated academic health system," npj Digital Medicine (October 26, 2026; early version of the accepted paper), https://doi.org/10.1038/s41746-026-03326-6 (opened 2026-11-03)
  7. Bair H, "What Ambient AI Scribes Cannot Hear," Journal of General Internal Medicine (Perspective, October 29, 2026), https://doi.org/10.1007/s11606-026-10779-8 (opened 2026-11-03)
  8. ClinicalAIWeekly, "Standalone AI Scribe or EHR-Bundled AI Notes: How to Choose" (blog, published 2026-10-29), https://www.clinicalaiweekly.com/blog/standalone-vs-ehr-bundled-ai-scribe (opened 2026-11-03)
  9. ClinicalAIWeekly, "Corrections policy and changelog" (Best Picks refresh and EHR incomplete-pricing flags, October 29; pilot methodology correction, October 30), (opened 2026-11-03)
  10. ClinicalAIWeekly, "Psychiatric EHR comparison 2026" (13 EHRs; TheraNest 74, CharmHealth 73, OptiMantra 73), https://www.clinicalaiweekly.com/practice-stack/ehr-comparison/ (opened 2026-11-03)

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.

Get The Behavioral Health AI Brief

One short, sourced email a week: what changed in AI scribe pricing, privacy terms, and features. Free, unsubscribe anytime.