Verified July 202613 min read

AI scribes for initial psychiatric evaluations

An initial psychiatric evaluation is the hardest test for an AI scribe. The best output in 2026 comes from Twofold Health, which structures HPI, past psychiatric history, social/family/substance history, MSE, risk assessment, DSM-5 differential, and plan as discrete sections rather than a flat SOAP. JotPsych and Freed are viable alternatives but require template customization to match Twofold's default structure.

The canonical initial-evaluation structure

  • Chief complaint and reason for referral.
  • History of present illness (HPI).
  • Past psychiatric history — including prior hospitalizations, trials, and response.
  • Past medical history and current medications.
  • Substance-use history.
  • Family psychiatric history.
  • Social and developmental history.
  • Mental status exam.
  • Risk assessment (suicide, homicide, other).
  • DSM-5-TR differential diagnosis with primary and rule-outs.
  • Formulation and initial treatment plan.

What breaks generic scribes on intakes

Generic scribes collapse the eleven-section structure into a SOAP or medical H&P template. The past psychiatric history, MSE, and risk sections lose their internal structure; the differential is often reduced to a single leading diagnosis; and social/developmental history is folded into 'Subjective.' Fixing this by hand takes 15–30 minutes per intake — enough to offset most of the scribe's time savings.

Field performance on 60-minute intakes

Twofold Health

Preserves all eleven sections by default. Lowest post-hoc editing time in our test.

JotPsych

Psychiatry-labeled with intake templates; structure held for HPI and MSE but differential was inconsistent.

Freed

Excellent transcription and prose; requires a custom intake template to match psychiatry-native structure.

Commure Scribe

Enterprise generalist; workable on Epic but weaker on psychiatric-history depth.

Pilot checklist for intakes

  • Run at least three real 60-minute intakes.
  • Measure editing time per note.
  • Verify past psychiatric history is preserved with structure, not collapsed.
  • Check that differential includes rule-outs, not just a primary.
  • Confirm risk-assessment language is verbatim.

Frequently asked

Can an AI scribe write a full 60-minute intake?
Yes. Twofold Health, JotPsych, and Freed are all validated at 60+ minutes of continuous capture in our 2026 testing.
How much editing time should I expect on an intake?
Two to eight minutes for the best products, fifteen or more for scribes that flatten psychiatric structure into SOAP.
Do AI scribes generate DSM-5-TR differentials automatically?
They surface the diagnoses discussed and structure them as differential when the visit does — they do not invent diagnoses you did not consider.

Scribes referenced in this guide

Related guides