Dictation vs. recording sessions for AI therapy notes

The most common reason therapists give for not using an AI scribe has nothing to do with technology: “My clients don’t want to be recorded — and honestly, I’m not sure I want to record them either.”

That instinct deserves respect, not a workaround. A therapy session is a protected space, and a visible recorder changes it. Some clients will consent and genuinely not mind. Others will consent and then self-censor for fifty minutes. A consent form can’t fix the second case, because the client signed it.

The good news: recording was never a requirement for AI-assisted notes. There are two workflows, and the one almost nobody markets is often the better clinical fit.

Workflow one: record the session

The default pitch. Audio of the full session is captured, transcribed, and summarized into your note format.

What it’s genuinely good at: accuracy and completeness. The note draws on what was actually said, with quotes available. Nothing depends on your end-of-day memory. If you see eight clients back-to-back, session five doesn’t blur into session six.

What it costs: the recorder is in the room — clinically and relationally. You need informed consent (and should want it to be genuinely informed: where the audio goes, how long it’s kept, who processes it). With cloud scribes, that consent conversation gets complicated, because the honest answer to “who else hears this?” involves servers, subprocessors, and retention policies. With on-device processing the answer is shorter — “it stays on this computer and is deleted when I finish the note” — but the client’s comfort with being recorded at all is still the threshold question, and they’re allowed to say no.

Workflow two: dictate after the session

The client is never recorded. After the session — between clients, or at day’s end — you speak a two-to-three-minute clinical summary in your own words: what the client reported, what you observed, what you did, how they responded, what’s next. The AI structures your dictation into your note format.

What it’s genuinely good at: consent simplicity and the therapeutic frame. There’s nothing in the room but the two of you. The only voice ever processed is yours. Clients who would decline recording — trauma histories, paranoid presentations, teenagers, anyone in a custody dispute — lose nothing. And the act of dictating is itself a brief structured reflection on the session, which some clinicians find sharpens their formulation rather than just documenting it.

What it costs: the note is bounded by what you remember and choose to say. No verbatim quotes, no catch of the phrasing you didn’t register in the moment. It adds two or three minutes per client of actual dictating — though against the 10–15 minutes of typing it replaces, that math still wins decisively.

Choosing between them (you don’t have to)

This isn’t a permanent fork. The sensible unit of choice is per client, per situation:

New client, trust still forming → dictate.
Long-term client comfortable with recording, complex session content → record.
Client explicitly asks not to be recorded → dictate, and document the preference.
High-stakes documentation (risk assessments, mandated treatment) → record if consented; the verbatim record protects everyone.
Couples and family work → the consent bar is every participant; when in doubt, dictate.

A scribe tool should treat both as first-class. (In CouchNotes, dictation mode is a guided flow — prompts walk you through the sections of your note format, you re-record any segment, and the draft lands the same way a recorded session’s would. It exists because the recording-averse case is common, not an edge case.)

What to say to clients, either way

Whatever workflow you use, the disclosure conversation is the same three sentences:

What happens: “I use software on my computer to help me draft my session notes.”
What it touches: either “with your permission, sessions are recorded and the recording is deleted once my note is done” — or “you’re never recorded; I dictate a summary afterwards in my own words.”
Their control: “You can say no, or change your mind anytime, and it won’t affect your care.”

If your tool processes audio in the cloud, sentence two needs to say so honestly — “the recording is processed on the company’s servers” — and you should be ready for the follow-up question. If processing is on-device, the airplane-mode test makes the claim demonstrable rather than contractual.

We keep free, adaptable client consent and disclosure templates for both workflows — the recorded-session consent and the dictation-only disclosure. Take them, put them on your letterhead, run them past your counsel.

The principle underneath both: the client’s comfort is a clinical variable, not an obstacle to your tooling. Pick the workflow that protects the work; the documentation will follow.

CouchNotes ships both workflows — record, dictate, or import — with everything processed on your Mac. The waitlist gets beta invites and early-access pricing.