2026-08-26 · 4 min read

AI for therapists: writing help that never touches confidentiality

Why do therapists need double caution with AI?

Because their material is doubly sensitive: it identifies a person and exposes their inner world — mental state, crises, family relationships. Data-protection regimes worldwide treat mental-health information as a specially protected category, and professional ethics go further still.

What can AI still do for a therapist?

The writing around therapy, not the therapy itself: turning bullet points into a structured session note, drafting a referral letter, organizing a case summary for supervision, or softening the language of a letter to parents. All of it works when the client appears as PERSON_001.

  • Session notes — turning raw bullets into a structured, professional note.
  • Referral letters — to a physician, school or insurer, in language fit for the recipient.
  • Supervision prep — organizing a case description for consultation, with no identifying details.
  • Sensitive wording — phrasing a recommendation or assessment precisely and kindly.

What identifies a client even without a name?

In therapy, context identifies as strongly as a name: age plus town, a rare profession, an unusual family event, dates and institutions. A note about 'a 40-year-old from a small town who lost his brother in an accident' is identifiable with no ID number at all.

  • Name, ID and phone number — the obvious part.
  • The combination of age, town and occupation — especially in small communities.
  • Unique events: an accident, a legal case, an admission to a specific facility.
  • Family members' names, the children's school, the workplace.

What does a safe workflow look like in a practice?

  1. One iron rule: no text about a client enters an AI tool before anonymization.
  2. Replace the name with a consistent token — PERSON_001 — and do the same for family members, town and institutions.
  3. Blur telling details: 'a loss in the family' instead of the specific story, when the detail isn't needed for the wording.
  4. Clean the file's metadata — the Author field and comments expose names even when the text is clean.
  5. Restore the real details only in the client's record, on your own machine.

The principle is the same one guiding clinicians who use AI: the tool gets the professional content, never the identity. The general list of what should never be pasted into a chatbot applies here with extra force.

Frequently asked questions

May a therapist use ChatGPT at all?

Yes — on text that doesn't identify a client. Confidentiality forbids exposing client information, not using a writing tool. An anonymized note can be processed safely.

Is removing the name enough?

Usually not. Age, town, occupation and unique events identify a person without any name, especially in small communities. Every telling detail needs replacing or blurring, not just the name.

Can I paste a case description for consultation?

Yes, under the same rules as presenting a case in supervision: no identifying details. Consistent anonymization lets you get help organizing and phrasing without exposing the client.

Does HIPAA prohibit using AI tools?

HIPAA restricts sharing identifiable health information outside defined rules and agreements — which a public AI chat doesn't meet. De-identified text is a different matter, which is why anonymizing first is the practical path.

Does a dedicated clinical-documentation AI tool solve this?

It improves the contractual terms, but the data still leaves your practice to an external vendor. Anonymizing before the paste protects independently of any vendor — and stays under your control.