The objection is rarely about accuracy. It is that the session audio has to leave the room to be processed, and a therapy hour is not ordinary material to send somewhere.

Five reasons come up repeatedly. The audio goes to a third party the client never chose. Consent becomes a live conversation at the start of the therapeutic relationship, and some clients say no. The note comes back sounding like somebody else, and the reasoning that made it a clinical document has been replaced by a summary of what was said. Writing notes is itself a thinking process, and outsourcing it changes what you notice. And the recording exists — even briefly — which is a record, a retention question, and a breach surface that a handwritten note never was.

The honest counterweight is that the time saved is real. Practitioners who use these tools describe getting evenings back, and burnout from documentation load is not a small problem. An argument against these tools that leaves that out is not one worth acting on.

Which objections hold up best?

Where the audio goes is the strongest, and it is answerable in principle. Processing that happens on hardware the vendor controls, in a jurisdiction you can name, is a different proposition from audio sent to a general-purpose service. The question to ask is not "is it secure" but "what specific systems does the audio pass through, in which countries, and who is contractually allowed to see it."

Consent is next, and it's not a formality. A client being asked in their first session whether their words can be processed by software is being asked something meaningful, and their answer sometimes shapes what they say for the rest of the year. Some practitioners find that cost alone decisive.

The record surface holds up too. Whatever is created — audio, transcript, draft note — is potentially part of the clinical record, with retention and disclosure consequences. If it's held anywhere, someone can eventually be asked for it.

Which ones are weaker than they sound?

"It writes bad notes." Sometimes true, sometimes not, and improving. If this is your only objection, it is an objection to a tool rather than to the category, and it will age.

"I'd never trust it with something clinical." Most practitioners already trust third-party software with the chart, the schedule, the email and the video. The scribe is different in degree and in what it processes, but "I keep software out of my clinical work" is usually not accurate as stated.

"It's not compliant." A word doing a lot of work with very little behind it. Ask what the vendor actually publishes: where data is held, who sub-processes it, whether audio is retained, and whether anything is used to improve their models. The answers vary enormously between vendors, and some of them are good.

What about the deskilling argument?

Take it seriously, because it's the one that isn't about data at all.

Writing a note is where a lot of practitioners do their thinking. You put down what happened and notice a pattern you missed in the room. You write the formulation line and realise you don't believe it. That work happens in the writing, not before it.

A generated draft you edit is a different cognitive act from a note you compose. Not necessarily worse — editing forces a review that a hurried note never gets — but different, and the difference compounds over a career. It is particularly worth thinking about early in training, when the writing is doing the most.

What do clients think?

Ask, because the assumption runs both ways.

A growing number of clients now assume everything is being recorded and processed by something, whether or not it is. For some, being told plainly that the session is not recorded is a relief they didn't know they were waiting for. For others it is genuinely a non-issue, and being asked feels like a formality.

What's not reasonable is deciding on their behalf. If a scribe is running, the client knows before it starts, in words, with a real option to decline that doesn't cost them anything.

Is there a middle position?

Several, and most practitioners who have thought about this land in one.

• Use it for some sessions and not others. Intake, yes; trauma processing, no.

• Use it only where the processing is local. Software running on your own machine, or a vendor's own hardware in a country you can name, changes the first objection substantially.

• Use it and write your own reasoning. Let it summarise content and write the clinical judgement yourself — which is the part that matters in the record and the part it can't do anyway.

• Don't use it, and fix the scheduling instead. Ten minutes between sessions solves a large fraction of the documentation problem for free.

What should I ask a vendor?

Five questions, in writing, before anything runs in a real session:

1. Where is the audio processed, and in which country?

2. Is the audio retained after the note is generated, and for how long?

3. Which sub-processors touch it? Names, not categories.

4. Is any of it used to train or improve a model?

5. What happens to everything if I close my account?

A vendor who answers all five clearly has told you a lot. One who answers with a compliance badge has told you something too.

The short version

The objection worth holding is about where the audio goes and what remains afterwards, not about whether the writing is any good. Ask the five questions, decide per session type rather than in the abstract, and if you decline, decline for the reason that actually applies rather than for a general unease — you'll be asked to explain it eventually.

About Rivet

Rivet is a Canadian practice line built for therapists — a separate number for calls, texts and voicemail, with video sessions and clinical tools in the same place. Your data stays in Canada, and transcription runs on Rivet's own hardware rather than a third-party AI service.

One plan, $65 CAD a month, everything included. Fourteen-day trial, no card.

See what's included · Start the trial