Better notes, fewer interventions — a paradox worth paying attention to
STUDY: Castro VM et al, JAMA Psychiatry 2026;83(3):281–286
STUDY TYPE: Matched retrospective cohort study
FUNDING: National Institute of Mental Health
Background
AI ambient scribes are now standard-issue in many practices, largely sold on the promise of reducing documentation burden. But better notes don’t automatically mean better care — and this study tests whether that gap is real.
The Study
Researchers analyzed over 20,000 outpatient notes from two large academic health systems in Massachusetts, comparing AI-scribed visits against three controls: human-scribed, contemporaneous unscribed, and pre-deployment unscribed. Groups were matched on age, sex, race, and prior depression diagnosis. The notes were from primary care, but the study focused on mental health treatments.
AI-scribed notes documented more psychiatric symptoms across mood, cognition, social function, and more. But visits using AI scribes were significantly less likely to result in a depression-related intervention (diagnosis, antidepressant prescription, or referral): 14% vs. 17% in both comparator groups. In a multivariate model, AI scribe use was associated with a 17% lower odds of any psychiatric intervention (adjusted odds ratio 0.83). Human scribes showed no such effect.
Limitations: Retrospective study with confounding risks — AI scribes weren’t randomly assigned, and unmeasured differences between practices could explain some of the gap. Still, the effect was specific to AI scribes; human scribes didn’t show it.
Practice Implications
- To explain this seeming paradox, the authors raise the “autopilot” hypothesis — that automating documentation may make clinicians less active overall, not just less burdened. It’s plausible. When the note writes itself, the clinician may disengage from the material it contains.
- On the other hand, AI scribes help psychiatric clinicians maintain eye contact and empathic engagement, potentially improving decision-making and outcomes. Unfortunately, this study doesn’t show that.
—Chris Aiken, MD
Director, Psych Partners
Editor in Chief, Carlat Psychiatry Report
What’s Your Take? Share in Comments
- Are AI Scribes helping or harming treatment planning?
4 comments
Sarah Franklin
April 5, 2026 at 7:51 am
When it comes it eye contact and engagement, how about we just don’t concurrently document during our patients appointments? I never have during follow-ups, and have been told my charting is extremely detailed. Most of my referrals come from my patients.
Krystle
April 5, 2026 at 10:37 am
AI scribes should only assist with documentation. They are not helping if the providers using them are not doing their due diligence to ensure appropriate diagnosese are given, and appropriate medications and interventions are prescribed. Reading the notes that are taken, editing them, adding content, and removing errors are the responsibility of the provider. AI scribes can be helpful, but they should only be a tool we use, not a replacement for clinical assessment, diagnosis, and treatment. The application of our knowledge and experience is crucial in providing care.
Jeanette
April 5, 2026 at 1:38 pm
I see both sides of that in my practice and intentionally maintain critical thinking and engagement in the therapeutic process. I always review my notes and edit them, because AI doesn’t always get it right. It’s a tool to leverage my engagement with my patients not replace or lessen it.
Sarah
April 8, 2026 at 1:16 pm
Remember – with AI , YOU and YOUR PATIENTS are (wittingly or unwittingly) being farmed for data. Tech companies don’t care about patients – they want your data.