AI documentation tools aren't one-size-fits-all. Here's how to use AI to generate accurate therapy notes for CBT, DBT, and EMDR — and what to look for in a tool that supports your clinical approach.
AI therapy notes are only as modality-specific as the input you provide. A tool that accepts structured session data — including what CBT techniques were used, which DBT skills were practiced, or what EMDR processing phase was reached — can produce draft notes that reflect your clinical approach. A generic input produces a generic note.
The most common concern clinicians raise about AI documentation is not privacy or accuracy in the abstract. It is specificity. Will the AI understand that this was a DBT skills session focused on distress tolerance, not a generic supportive therapy session? Will it capture that the patient's SUD went from an 8 to a 3 and that the blocking belief shifted?
The answer depends on two things: the input you provide and the tool's ability to use it. This post breaks down what AI documentation looks like for each of the three most common structured modalities — CBT, DBT, and EMDR — and how to get notes that actually reflect what happened in the room.
Different therapeutic approaches capture fundamentally different clinical data.
A CBT session centers on cognition: what automatic thoughts the patient identified, which distortions were challenged, what behavioral experiments or homework assignments were assigned or reviewed. A DAP note that says "Patient discussed negative thoughts about work" misses everything clinically meaningful.
A DBT session centers on skills: which module was the focus, what the patient practiced between sessions, what the diary card showed, whether a chain analysis was completed. Progress in DBT is measured by skill acquisition and use — documentation that doesn't capture those elements doesn't capture the therapy.
An EMDR session has a structure built around processing phases, measurable distress ratings, and tracked shifts in cognition. The SUDs number at the start and end of a session is not administrative data — it is the primary outcome measure. Documentation that omits it is clinically incomplete.
Generic AI notes fail not because the AI is incapable, but because the input was generic. The fix is structured input that tells the tool what happened clinically, not just that a session occurred.
The more specific you are, the more accurate the draft. Include:
A session input like "Reviewed thought record from last week — patient identified catastrophizing around work performance review. Challenged with Socratic questioning, patient acknowledged evidence against the feared outcome. Assigned behavioral experiment: attend next meeting without preparing responses in advance, log actual vs. feared outcome" produces a CBT-accurate note. A session input like "Discussed work stress and cognitive patterns" does not.
DBT sessions are structured enough that the session flow itself is documentable:
Chain analyses in particular benefit from structured input. A good input captures the precipitating event, the key links (thoughts, feelings, behaviors), and the solution analysis outcome. That level of specificity lets the AI produce a note that actually reflects the DBT model rather than a paraphrased summary of what was talked about.
EMDR sessions have measurable outcomes that belong in the note. Include:
Because EMDR has a defined phase structure, AI tools can use the phase as a template anchor — what was assessed in the assessment phase, what SUD tracking occurred in desensitization, what VOC rating closed out installation. Without knowing the phase, the note has no structure to organize around.
The note format you use matters as much as the content. The same session can produce very different documentation depending on whether it is organized as a SOAP note, a DAP note, a BIRP note, or a narrative summary.
| Format | Structure | When it works best |
|---|---|---|
| SOAP | Subjective / Objective / Assessment / Plan | Standard clinical settings, EHR-integrated workflows |
| DAP | Data / Assessment / Plan | Many behavioral health practices, streamlined format |
| BIRP | Behavior / Intervention / Response / Plan | Insurance-facing documentation requiring clear intervention justification |
| Narrative | Prose summary | Consultation notes, supervision documentation, some specialty settings |
A good AI documentation tool lets you specify the format per session — and ideally lets you combine format with modality context so that the Assessment section of a SOAP note reflects CBT cognitive outcomes, not generic clinical impressions.
If a tool locks you into one format regardless of clinical context, you will spend time reformatting every note to match your practice's requirements. Format flexibility is not a luxury feature — it is a baseline expectation.
The best time to configure modality-specific documentation is before the first session, not after the third — and the simplest way is to stop re-typing the context.
In PsyFiGPT, a saved prompt can state your modality, the note format you want, and the elements you always want captured ("CBT-focused DAP note; always include distortions identified, homework reviewed, homework assigned"). Team prompts let a group practice share that structure across clinicians while each keeps their own voice. From then on you paste your session input — or pull in a recorded transcript from PsyFi Notes — and the prompt supplies the framework.
If you record sessions, set it at capture time instead: the recorder's Session details card has a Modality field and a Note format dropdown (SOAP, DAP, BIRP, GIRP, or Narrative, with a default you set once in Preferences), so the draft is organized around your approach before you ever open it.
| Modality | Essential input elements |
|---|---|
| CBT | Cognitive distortions identified, techniques used, homework reviewed and assigned, patient's cognitive shifts |
| DBT | Module and skill(s) focused on, diary card summary, chain analysis if completed, skill use between sessions |
| EMDR | Phase worked in, target memory or NC/PC, starting and ending SUD/VOC, processing themes, closure method if applicable |
| All modalities | Note format desired (SOAP/DAP/BIRP/narrative), patient response, plan for next session |
The more structured your input, the less editing the draft requires. Most clinicians find that after a few sessions, they develop a consistent input rhythm — a 2-3 minute voice note or structured prompt after the session that gives the AI enough to produce a draft that is 80-90% ready to sign.
PsyFi Notes records the session (in the room or over telehealth) and returns a speaker-attributed transcript with a draft in the format you chose — SOAP, DAP, BIRP, GIRP, or narrative — tagged with the modality you set on the recording. For deeper modality-specific work, pull that transcript into PsyFiGPT, which is built for behavioral health documentation specifically: the prompting system understands clinical structure — CBT thought records, DBT skills modules, EMDR processing phases — and can organize a draft around what actually happened in the session rather than producing a generic clinical narrative.
The practical difference: a PsyFiGPT note for a DBT chain analysis session names the target behavior, traces the chain links, and documents the solution analysis. It does not summarize the session as "patient discussed a behavioral incident and identified contributing factors."
That specificity comes from the session input you provide and from a tool designed to use clinical context — not just generate plausible-sounding notes.
If your AI documentation tool isn't capturing your clinical approach, the issue is usually input structure, not the tool's ceiling. Try it with your next session and compare the draft to what you would have written yourself — or contact us to talk through your specific modality.
PsyFi Notes is HIPAA-aligned clinical documentation software backed by a Business Associate Agreement. For questions about modality-specific documentation in your practice, contact our onboarding team at [email protected].
