Not with identifiable client details on a personal ChatGPT plan. Therapists can use ChatGPT for session notes only in narrow ways: on a business plan with a signed data-processing agreement and your professional body's requirements met, or with genuinely de-identified shorthand. For most solo practitioners, a therapy note tool with proper data terms is the safer choice.
The reason is that therapy content identifies people by its nature. Remove a client's name and you are still left with a bereavement, a particular workplace conflict, a relationship breakdown, an age and a job. In a small community, any two of those can point to one person. And ChatGPT has a second habit that matters in therapy: it adds interpretations you did not make, in confident clinical language, which then sit in a record the client may one day read.
Four ways therapists try it, and where each lands
| What the therapist does | Verdict | Why |
|---|---|---|
| Pastes session details, with names, into a personal ChatGPT plan | No | No data agreement, training switch on by default, content sits in chat history |
| Types de-identified shorthand into a personal plan to get a structured draft | Risky, limited | Hard to de-identify therapy content fully; still no agreement covering it |
| Uses ChatGPT Business with a signed data-processing addendum and settings reviewed | Possible, with conditions | No training by default and a contract, but you must still meet health-data and professional rules |
| Uses a note tool built for therapy with a data agreement | Usually the best fit | Designed for clinical records, with retention and consent settings built in |
The rest of this tutorial explains each row, then covers the uses of ChatGPT that never touch a client's session at all, which is where most therapists get real value from it.
What happens to what you type into a personal plan
On the Free, Go, Plus and Pro plans, the defaults are built for individuals, not for clinical records:
- Training. The setting "Improve the model for everyone", under Data controls in Settings, is on by default. Unless you switch it off, your conversations can be used to train OpenAI's models. Switching it off is not retroactive.
- History. Everything you type stays in your chat history until you delete it, visible to anyone who opens your account on a shared laptop or phone.
- Memory. If memory is on, ChatGPT can carry details from one chat into later ones. A client detail from a notes chat can surface in an unrelated conversation. Check your memory settings and what is stored.
- Temporary Chat. Temporary chats stay out of history and memory and are not used for training, but OpenAI keeps them for up to 30 days.
- No agreement. There is no contract between you and OpenAI that covers client health information on a personal plan.
An illustrative mistake that shows how this goes wrong in real life: a therapist used her personal ChatGPT account on the family laptop to tidy notes, with first names only. Her teenage son opened ChatGPT to do homework, scrolled the sidebar and saw a chat titled "Notes: Dan, panic attacks, marriage". Nothing was hacked. The history simply did what history does.
Why removing the name isn't enough
De-identifying ordinary business data is fairly mechanical: strip names, numbers and addresses. Therapy content is harder, because the story is the identifier. Here is an illustrative before and after.
"De-identified" the usual way: "Client (M, 34, firefighter) discussed guilt about his brother's death in the warehouse fire in March. Conflict with his station manager about returning to shifts. Wife considering leaving."
No name, but a 34-year-old firefighter whose brother died in a warehouse fire in March is identifiable to anyone in that town, and to the client himself if he ever saw it.
De-identified properly: "Client discussed guilt following a family bereavement. Work stress around a return to duties. Relationship strain at home. Sleep reduced. Practised grounding technique."
The second version is safe to share far more widely, but look at what it gives ChatGPT to work with: very little that you could not have written yourself in the time it took to strip the first version down. That is the practical problem. The more properly you de-identify therapy notes, the less there is for AI to help with.
If you do use ChatGPT for drafting: a narrower workflow
Some therapists do use ChatGPT for the structuring step only, with properly de-identified shorthand, on a plan whose settings they have reviewed. If you choose to, keep it tight:
- Write your own shorthand notes first, in your records system. That is the record.
- Make a second, de-identified version: no names, ages, jobs, places, dates, or unique events.
- Use a prompt that forbids interpretation (below).
- Paste the result back into your records system, edit it, and delete the chat.
- Never paste anything a client wrote to you, such as messages or journal entries.
Turn these shorthand therapy notes into a progress note using these
headings: Session focus / Interventions used / Client response /
Risk / Plan.
Rules:
- Use only what is in the notes. Do not add diagnoses, attachment
styles, formulations, defence mechanisms or clinical interpretations.
- Do not add risk information. If risk is not mentioned, write
"Risk: [therapist to complete]".
- Plain, neutral language a client could read without feeling judged.
- Past tense, under 150 words.
Notes: bereavement guilt, work stress re return, relationship strain.
sleep 4-5 hrs. grounding practised, found helpful. homework: sleep
diary. next: continue grief work.An illustrative output, and what the therapist removed:
Session focus: Guilt following a bereavement, work stress about returning to duties, and strain in a close relationship.
Interventions used: Grounding technique practised in session.
Client response: Client reported the grounding technique as helpful. Client's guilt may reflect unresolved complicated grief.
Risk: [therapist to complete]
Plan: Continue grief-focused work. Client to keep a sleep diary.
The italic sentence is an interpretation the therapist never made, despite the prompt forbidding it. "Complicated grief" is a clinical concept with a specific meaning. It came out. This is the most common failure, and it is why the draft is never pasted in unread. Note also that the reduced sleep dropped out of the draft entirely, so the therapist added it back.
The interpretation problem, and why it matters more in therapy
General chat assistants have read a great deal of therapy writing and like to sound like it. Given plain notes, they reach for formulation language: "presents with an anxious attachment style", "displays avoidant coping", "exhibits cognitive distortions". Some of those phrases may be close to what you think. None of them should be in a record unless you decided to put them there.
It matters for three reasons. Clients in many places can request their records, and reading "exhibits cognitive distortions" about yourself lands very differently from "we talked about how thoughts affect mood". Records can be requested in legal proceedings, where a confident AI phrase becomes your professional opinion. And another clinician picking up the case will take the note at face value. Write the prompt to forbid interpretation, and read the output looking for it anyway.
What ChatGPT Business changes, and what it doesn't
ChatGPT Business costs $25 per user a month billed monthly, or $20 billed annually, with a minimum of two seats, so a solo therapist pays for two: about $40 to $50 a month. In return, OpenAI does not train on business content by default, you get admin controls, and OpenAI says it can sign a data processing addendum for ChatGPT Business, Enterprise and its API. There is more on the plan differences in ChatGPT Plus versus ChatGPT Business.
What it does not change:
- It is not a clinical records system. You still need your notes in a proper record with the access, retention and deletion rules your profession expects.
- It does not settle whether health-data rules where you practise allow this use. Some require specific contract terms for health information; ask your data-protection adviser or professional indemnity provider.
- It does not meet your professional body's expectations automatically. Several have published guidance on AI; check yours.
- It does not remove the need to tell clients. That question has its own answer in whether therapists should tell clients they use AI.
OpenAI has also launched a free clinician edition of ChatGPT, but at launch it was limited to verified clinicians in certain medical professions in one country, and therapists were not on the core list. Check OpenAI's clinicians page if you want to know whether that has changed for you.
Therapy note tools compared with ChatGPT
Purpose-built therapy note tools do the same drafting job with the clinical plumbing included: data agreements, retention settings, consent prompts, and templates for progress notes. Examples include SimplePractice's Note Taker (an add-on at $35 per clinician a month), Upheal and Mentalyc. Many also offer a mode that works from your typed or dictated summary rather than a session recording, which suits therapists who do not want sessions recorded at all.
They are not risk-free either, and defaults move. In June 2026 SimplePractice changed its Note Taker so that new users are opted in by default to keeping de-identified transcripts. That is not necessarily a problem, but it is exactly the kind of setting a therapist should know about and choose, rather than inherit. Whatever tool you consider, ask: where is session content stored, for how long, is any of it used to improve the product, can you switch that off, and will the vendor sign the agreement your rules require. The full list is in the confidentiality checklist for AI note-taking tools.
A quick cost comparison for a solo practitioner seeing 20 clients a week, illustrative and at list prices: ChatGPT Business at two seats runs about $40 to $50 a month and still leaves you writing the prompt, stripping identifiers and pasting results around. A therapy note tool at around $35 a month fits into the records system and handles the data terms. For most solo therapists the dedicated tool is both cheaper in time and cleaner on confidentiality.
A middle path: dictate a summary, don't record the session
Many therapists do not want sessions recorded at all, for good reasons: clients speak less freely, and a recording is one more thing to protect. There is a middle path that still saves time. Straight after the session, spend two minutes speaking a summary into a therapy note tool's dictation mode, and let it structure the progress note.
An illustrative two-minute summary: "Session eight. Focus on the bereavement guilt again, and worry about going back to work next month. Practised grounding, which she said helped. Sleep still poor, around four to five hours. No change in risk; she denied any thoughts of harming herself when I asked. Homework is a sleep diary. Next week, continue with the grief work and look at the return-to-work plan."
That gives the tool your words, in your order, with your judgement about risk stated plainly. There is nothing for the AI to misattribute, because only you are speaking, and far less for it to interpret, because you have already said what mattered. The resulting draft usually needs a light edit rather than a hunt for invented content. It is also easier to explain to clients: "I dictate a short summary after we finish" is a very different sentence from "an AI listens to our sessions".
Couples, family and group work make recording harder still, because every person in the room has their own confidentiality and their own right to decline. Dictating a summary afterwards sidesteps the problem of one partner agreeing to recording and the other feeling unable to say no.
Uses of ChatGPT that never touch a client's session
This is where most therapists get safe, everyday value, on any plan, because no client information goes in:
- Note templates. "Draft a progress note template for 50-minute sessions using a person-centred approach, with headings and prompts under each." You fill it in your records system.
- Psychoeducation handouts. A one-page sheet on sleep hygiene, the stress response, or grounding techniques, written at a reading level your clients find easy. Check every claim before handing it out.
- Letters and forms. A structure for a letter confirming attendance, a contracting agreement, or a cancellation policy.
- Practice admin. Website copy, a waiting-list email, a fee-change announcement.
- Your own learning. Summaries of approaches you are reading about, questions to take to supervision, CPD planning.
An illustrative handout request: "Write a one-page handout explaining the 5-4-3-2-1 grounding technique for adults experiencing anxiety. Warm, plain language, short paragraphs, no clinical jargon, and a note to talk to their therapist if it doesn't help." The draft took seconds, and the therapist's edit was to remove a line claiming the technique "stops panic attacks", which overstated it. More ideas of this kind are collected in safe AI uses for counsellors that never touch client notes.
Answering the question for your own practice
Three questions settle it for most therapists:
- Does any client information need to go into the tool? If no, as with templates, handouts and admin, ChatGPT on a plan with the training switch off is fine.
- If yes, is there a signed agreement that covers health information, and do your professional body and health-data rules allow it? If you cannot answer yes to both, stop there.
- If both are yes, is a general chat assistant still the best tool? Usually a therapy note tool is simpler and safer, because it was built for this job.
Whatever you decide, write it down: which tool, which plan, which settings, what goes in and what never does, and how clients are told. That one page is what you show a supervisor, an insurer or a client who asks, and it is far easier to write before a problem than after one.
An illustrative version, filled in by a solo counsellor:
- Tool for notes: therapy note tool in dictation mode only; no session recording. Data agreement signed; transcript retention set to the shortest option; product-improvement sharing switched off.
- ChatGPT: personal Plus plan, training switch off, memory off. Used only for templates, handouts, website copy and my own learning. No client information, ever, including first names.
- Clients told: in the contracting agreement and verbally at the first session; they can opt out of AI-assisted notes with no effect on their therapy.
- Reviewed: every six months, and whenever a tool changes its terms or defaults.
Six lines, and every question a supervisor is likely to ask has an answer.
Follow-up questions from therapists
Is Temporary Chat in ChatGPT safe enough for session notes?
It helps but does not solve the problem. Temporary chats do not appear in your history or feed memory and are not used for training, but OpenAI keeps them for up to 30 days, and you are still sending client material to a service without a data agreement covering it. Treat it as a way to reduce exposure for de-identified drafting, not as permission to use client details.
Can I use ChatGPT to write a treatment plan?
You can use it to draft a generic structure, such as headings, goal formats and example measurable objectives, without any client information. Filling that structure with a specific client's history, formulation and goals is clinical work that belongs in your records system or a purpose-built tool with the right data terms.
What if my supervisor or employer already uses ChatGPT for notes?
Ask what plan they use, whether a data agreement is in place, and what their professional body's guidance says. Common practice is not the same as safe practice. If you work for an organisation, its policy applies to you, and if there is no policy, raise it rather than assuming silence means approval.
Do clients have a right to see notes that ChatGPT helped draft?
In many places clients can request access to their records, and that includes notes you drafted with AI help. Write every note as if the client will read it. That is another reason to strip out any interpretation the AI added that you would not stand behind in a conversation with the client.
Further reads
- ChatGPT Temporary Chat: When Your Staff Should Use It — What Temporary Chat does and does not protect.
- Can Solicitors Use ChatGPT Without Breaching Confidentiality? — How another confidentiality-bound profession answers the same question.
- Patient Data and AI: A Confidentiality Checklist for Small Practices — A confidentiality checklist for small practices using any AI tool.
- What Is an AI Scribe and How Does It Work in a Consultation? — How AI scribes record and draft notes, stage by stage.
- AI Session Notes and Follow-Ups for Coaches: A Weekly Workflow — How coaches, with lighter duties, handle AI session notes.
- Is ChatGPT Safe for Business Use? Risks, Settings and Plan Choice — The wider picture on ChatGPT's safety for business use.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help pages on data controls and Temporary Chat; OpenAI pricing for ChatGPT Business; OpenAI enterprise privacy information on data processing addenda; SimplePractice Note Taker pricing and default settings; published professional ethics guidance on AI in psychological practice.