Skip to content
Compliance

Are AI-generated therapy notes acceptable for Medicare and Medicaid billing?

CMS documentation rules for psychotherapy, what auditors actually look for in AI-drafted notes, and which scribes ship with billing-ready structure out of the box.

TherapyScribes Editorial10 min · 709 words
Reviewed by TherapyScribes EditorialUpdated Facts verified Methodology

The short version Yes — CMS does not prohibit AI-generated draft notes. What matters is that the billing clinician personally reviews, edits, and signs each note, and that the signed note supports the CPT code billed. The AI drafting the first version is invisible to the auditor; the signed note is what's on the record.

The practical risk with AI-drafted notes is not compliance in principle — it's auditor-visible tells that a note was accepted without meaningful review: identical phrasing across sessions, generic risk language, missing time-in/time-out for time-based codes, and thin medical-necessity language. Auditors have started flagging these patterns in 2025–2026 CMS reviews.

**For solo and small-group US therapy practices billing Medicare/Medicaid we currently recommend Twofold — its therapy-native templates ship with the fields Medicare auditors actually check (time-in/time-out for 90832/90834/90837, medical necessity language tied to the treatment plan, and Golden Thread continuity that shows progression across sessions). Mentalyc is the deeper option for CMHCs and larger practices with wide modality mix. Eleos Health** is the honest pick for CCBHCs specifically.

What CMS actually requires in a psychotherapy note Per the Medicare Benefit Policy Manual and MLN Matters guidance, a billable psychotherapy note must include:

1. Date, start time, and stop time (for 90832 / 90834 / 90837 — these are time-based codes, and auditors will deny without explicit times). 2. CPT code billed and rationale (why this code, not a shorter/longer one). 3. Diagnosis using current ICD-10 codes, tied to the treatment plan. 4. Type of therapeutic intervention used in the session (CBT, DBT, EMDR, IFS, motivational interviewing, etc.). 5. Client response to the intervention. 6. Medical necessity — why continued treatment is warranted, tied to functional impairment. 7. Plan for next session. 8. Signature and credentials of the billing clinician.

Every therapy-first AI scribe on our tested list can produce all eight fields. The differentiator is whether it does so *by default* in a way that survives a CERT or RAC audit.

Audit-visible tells to avoid - Verbatim phrasing across sessions. Auditors run text-similarity checks. If every note says "Client presented with congruent affect and appropriate hygiene" word-for-word, that's a flag. Edit these lines. - Generic risk language. "No SI/HI reported" on every note including the crisis session is a flag. Risk language should reflect what actually happened. - Missing time-in/time-out. If your scribe produces "Session length: 53 minutes" instead of "Start: 10:00, Stop: 10:53," add the times before signing. Time-based codes require explicit times. - Medical necessity that doesn't reference the treatment plan. "Client would benefit from continued therapy" is not medical necessity. Reference the specific treatment-plan goal and the client's current functional impairment. - Copy-paste of the assessment section. Bring the assessment forward from the intake once; update it as the client changes. Verbatim carry-over across dozens of sessions is a flag.

Billing-readiness of the major scribes

ToolTime-in/time-out by defaultMedical-necessity language tied to planGolden Thread across sessionsCPT suggestion
TwofoldYesYesYesSuggests, requires confirmation
MentalycYesYesPartialSuggests
UphealYesPartialPartialSuggests
Eleos HealthYesYes (CCBHC-oriented)YesSuggests
BlueprintPartialNoNoNo

What to change in your workflow (any tool) 1. Never auto-accept a CPT code. Confirm it against actual session length. 2. Add or verify time-in/time-out before signing. Set a personal rule: no signature without explicit times. 3. Rewrite the risk-language sentence on every note, even if it's the same content. Auditors see edit history in some EHRs. 4. Update the medical-necessity line to reference the current treatment-plan goal, not last month's. 5. Keep an internal audit trail of which notes were AI-drafted (many EHRs let you tag). This is not currently required by CMS but is defensive good practice.

Recommendation - Default for solo/small US practices billing Medicare/MedicaidTwofold. Ships with the fields auditors check. - CCBHCEleos Health. - Wide modality CMHCMentalyc.

The scribe doesn't create audit risk. Signing without reading does.

See also: HIPAA compliance for AI scribes and SOAP vs DAP vs BIRP.

Continue reading