Can therapists use ChatGPT for notes?

For some products. OpenAI's help center lists the products that are eligible for HIPAA use under a Business Associate Agreement, and as of October 2026 they are: ChatGPT for Healthcare, ChatGPT Enterprise with a regulated workspace, ChatGPT FedRAMP, ChatGPT for Clinicians, and the API with modified data retention. The API BAA is self-serve but, in OpenAI's words, "requires an established history of API usage."

The everyday plans are not on that list. OpenAI states directly: "We do not offer a BAA for ChatGPT Business," and Free, Go, Plus and Pro do not appear among the HIPAA-eligible products. Without a BAA, pasting client information into those accounts means sending PHI to a vendor that hasn't agreed to protect it under HIPAA.

ChatGPT for Clinicians, launched in April 2026, lets individual clinicians sign a BAA in Settings, under Agreements. OpenAI says that "eligible clinicians currently include physicians (MD/DO), nurse practitioners (NP), physician assistants (PA), and pharmacists."

That means a psychiatrist or a PMHNP may qualify. Psychologists, licensed clinical social workers, professional counselors and marriage and family therapists are not on the list as of October 2026. If you are a therapist in private practice, the realistic routes to a BAA with OpenAI are an organization-level product or the API, which few solo clinicians will set up.

ChatGPT Health, launched in January 2026, is a consumer feature for people asking about their own health. OpenAI says it "is not intended for clinical or covered-entity use and does not offer a Business Associate Agreement." It is not a tool for writing client notes.

On consumer accounts, OpenAI says it may use your content to train its models unless you turn off "Improve the model for everyone." Temporary chats are not used for training but may be kept for up to 30 days for safety purposes, and deleted chats are removed within 30 days unless OpenAI needs them for security or legal reasons. Turning training off is sensible, but it does not create a BAA.

The American Psychological Association's ethical guidance on AI in practice (June 2025, updated December 2025) says tools must be usable in compliance with HIPAA, that clients should give informed consent, and that "AI should augment, not replace, human decision-making." Its separate guidance on evaluating AI scribes says a signed BAA between the provider and the scribe company is required, and mentions dictation as an alternative to ambient listening.

The American Counseling Association's recommendations for counselors (2024) also call for explicit informed consent and HIPAA-compliant platforms. NASW has said its Code of Ethics workgroup is recommending supplemental guidance on AI; as of October 2026 we found no formal NASW guidance document.

It is riskier than it looks. HIPAA's de-identification standard lists 18 kinds of identifiers, including dates and small geographic areas, and therapy narratives are full of details (a workplace, a family event, an unusual job) that can identify someone even without a name. If you can't say with confidence that a note is de-identified, treat it as PHI.

Using ChatGPT for work that involves no client information, such as drafting a practice policy or a newsletter, is a different matter and raises no HIPAA issue.

If you want AI to draft notes, use a product that signs a BAA for your plan: your EHR's built-in AI, a clinical scribe, or one of OpenAI's HIPAA-eligible products if you qualify. Get client consent where a recording is involved.

If what you really want is to stop typing, you may not need a language model at all. Sapience Med is push-to-talk dictation that runs entirely on your Mac or Windows PC: you say the note in your own words and it types them into your EHR. It doesn't use ChatGPT or any cloud AI, it doesn't write the note for you, and since no vendor receives your words there is no BAA to sign. See how to dictate DAP, SOAP and BIRP notes for what that looks like in practice.