AI Note-Taking in Therapy: Consent, Psychotherapy Notes, and Where the Line Is
Ask a therapist what they'd give up for an extra five hours a week and most will tell you they'd give up the notes. Documentation after a full day of sessions is where a lot of clinicians lose their evenings.
AI scribes are the most mature AI use case in mental health right now. Several are on the market and built for behavioral health, including Eleos, Upheal, and Blueprint, and EHR vendors like Kipu have added their own. We list them because they exist and you'll run into them. This site hasn't tested them, and naming them isn't an endorsement.
Because these tools sit inside the therapy room, the rules are tighter than anywhere else on this site.
How AI scribes work
Most work one of two ways. Ambient capture listens to the session, live or through telehealth, and drafts a note afterward. Summary mode lets the clinician type or dictate a few lines after the session and expands them into a structured note. Both produce a draft. The clinician reviews, edits, and signs.
Summary mode is lower risk. Nothing from the client's voice leaves the room. The clinician decides what goes in. If you're nervous, start there.
Consent comes first
Recording or transcribing a therapy session changes the room. Clients deserve to know, in writing, before it happens, and they deserve a real choice to say no without it affecting their care.
Illinois turned this into law. Under the Wellness and Oversight for Psychological Resources Act, a licensed professional can't use AI for supplementary support in therapy where the session is recorded or transcribed unless the client is told in writing that AI will be used and what for, and the client consents. The same law bars licensed professionals from letting AI make independent therapeutic decisions, communicate therapeutically with clients, or generate treatment plans without the professional's review and approval. Other states are moving. Recording consent rules also differ from state to state. Your counsel should check the states you practice in.
A good consent conversation is short and honest. What the tool does. Whether audio is kept. Who can see it. That they can say no today and change their mind later. Then the clinician turns it off when asked, and nobody makes it weird.
Psychotherapy notes are their own category
HIPAA gives extra protection to psychotherapy notes. Under the HIPAA definitions in 45 CFR 164.501, these are notes recorded by a mental health professional documenting or analyzing the contents of a counseling session that are kept separate from the rest of the medical record. They don't include things like medication, session times, test results, or summaries of diagnosis, treatment plan, symptoms, and progress. HHS explains that most uses and disclosures of psychotherapy notes need the patient's authorization in its FAQ on mental health information.
For substance use disorder programs covered by Part 2, the 2024 rule created a similar category called SUD counseling notes that need their own separate consent, according to the HHS Part 2 fact sheet.
Why does this matter for AI? Because a scribe that captures everything said in session, and stores the transcript, may be creating something more sensitive than a progress note. Decide with counsel what your tool creates, where it lives, and what category it falls in before you turn it on.
Review is not optional
AI drafts can be wrong in ways that sound right. They can attribute a statement to the wrong person in a couples or family session. They can smooth over a risk statement. They can add an intervention you didn't do because it usually appears in notes like yours. They can miss sarcasm completely.
A signed note is your clinical and legal record. Read every draft as if you'll be asked about it under oath, because someday you might be. If a vendor's workflow encourages one-click signing without reading, that's a red flag.
Where the privacy line sits
Safe with de-identified or aggregate data. Measuring documentation time across the practice. Tracking late notes. Building templates from blank structures. Practicing on fictional sessions written for training. None of that needs real client data.
Needs a BAA. Any tool that receives session audio, transcripts, dictation about clients, or draft notes. That vendor is a business associate under HHS's business associate guidance. Some vendors publish their BAA or accept it at signup. Read it. Look at how long audio and transcripts are kept, whether data is used to train models, who the subcontractors are, and what happens to your data when you cancel.
Don't automate. Signing notes. Writing treatment plans without review. Flagging risk instead of a clinician. Detecting a client's emotions or mental state as a basis for decisions, which Illinois specifically bars licensed professionals from allowing. Recording any client who hasn't consented.
Groups, families, and couples
Group sessions add a problem: every person in the room has a privacy interest. Some products advertise group note features that separate speakers. Before you use one, decide how you'll get consent from every member, what you'll do if one person says no, and how the tool keeps one client's words out of another client's chart.
Families and couples raise the same problem in a smaller room.
How to start
Pick one clinician who's comfortable with technology and skeptical by nature. Have them use summary mode for two weeks with real clients who consented. Then ambient mode with a few clients who are comfortable with it. Compare the notes against what they would have written. Track how many drafts needed a real correction, not just a word change.
If error rates are low and clients are comfortable, expand slowly. Revisit consent forms once a year, or any time the vendor changes how it handles data.
Then go read the vendor evaluation guide. Ask every question on it. For intake and scheduling, see AI for mental health practice intake.
The point
The note was never the therapy. If an AI scribe gives a clinician back their evenings and their attention in the room, that's a good trade, as long as the client said yes and the clinician still owns every word that gets signed.
Frequently asked questions
Do I need client consent to use an AI scribe in therapy?
Get it in writing, every time, before the session is recorded or transcribed. In Illinois, written notice and consent are required by statute when AI supports therapy where the session is recorded or transcribed. Other states have their own recording and consent rules, so ask counsel.
Is an AI scribe vendor a business associate?
If it receives, stores, or processes session audio, transcripts, or notes for your practice, yes. You need a signed Business Associate Agreement before any session data goes in.
Are AI-drafted notes psychotherapy notes under HIPAA?
It depends on how they're kept. HIPAA defines psychotherapy notes as a mental health professional's notes analyzing a counseling session that are kept separate from the rest of the record. Most progress notes going into the chart are not psychotherapy notes. Ask counsel how your workflow classifies them.
Can the AI write the treatment plan?
It can draft. A licensed clinician has to review, change, and approve it. In Illinois, the law prohibits AI-generated treatment plans or recommendations without review and approval by the licensed professional.
What should I ask about session audio?
Whether audio is stored, for how long, whether it's deleted after the note is drafted, whether any of it is used to train models, and who the subcontractors are. Get the answers in writing.
Use cases for this guide
- Ambient scribe for therapy progress notes: An AI scribe listens to a session, with client consent, and drafts a progress note the therapist reviews and signs.
- Draft treatment plans for clinician review: AI drafts a treatment plan from session content that the clinician edits and approves.
- Note assistant built into an addiction treatment EHR: AI inside the EHR turns recorded or dictated sessions into structured notes in the program's existing forms.
- Pre-session recap from prior notes: Before a session, AI pulls what was discussed last time and what the client planned to work on.
- Turn a clinician's short summary into a full note: The clinician types or dictates a few lines after a session and the tool expands them into a structured draft note.