A clinician opens ChatGPT, Claude or Copilot and starts typing. These are general-purpose AI tools, the kind anyone can sign up for and type into to ask questions, draft text and summarize. This worksheet lays out eleven decisions a center makes about that use, with the realistic options and the downside of each, and turns your answers into a draft policy you can edit. Skip any you are unsure about.
This is for general-purpose AI tools only: ChatGPT, Claude, Copilot, Gemini. These tools are being adopted quickly in both personal and professional settings, which is why it is worth deciding how they are used in the center.
It does not cover AI that comes built into something else or that the center buys as a service: tools that listen to sessions and write the note, teletherapy and triage companies, AI features inside your record system, or student use of chatbots. Each of those needs its own contract and its own consent language. If your center uses any of them, this draft is a first step rather than the whole answer.
Go through the questions one at a time, ideally with a few people from your center rather than on your own.
There is no recommended answer. Centers differ on what they think these tools should be used for, and on what their institution and their state already allow. Before you decide, have three things in front of you: your institution's AI policy, your state's law on AI in mental health care, and the guidance from your licensing board and from the APA, ACA, NASW or NBCC.
A draft policy at the bottom, assembled from your answers and editable before you copy it out. It is a starting point rather than a comprehensive policy, and it is most useful as a way to get staff talking about how AI is already being used in the center. Answers save in this browser, so you can stop partway and come back.
Check your state first. Illinois, Nevada and Utah have passed laws on AI in mental health care, and other states are following. Some uses this page presents as open choices may be restricted or prohibited where you are.
Have any policy you produce here reviewed by your institution's legal counsel and privacy office before you adopt it.
Anything still open is listed at the end, and each one links back to its question.
Assembled from your selections. Edit it here, then copy it into your manual.
Anything you have not decided appears as a bracketed placeholder, so nothing gets settled by leaving it blank.
Three sections appear in every draft regardless of your selections: prohibited uses, individual responsibility, and review. Delete them if your center intends something different, but delete them deliberately.
Show it to the staff it governs while it is still a draft and their objections can still change it. Then take it to your institution's privacy office and legal counsel, with your state's law on AI in mental health care in hand. Then to whoever formally adopts policy at your institution. Set the next review date before you adopt it, because the tools this governs change faster than a manual does.
Comments, disagreements, and decisions I have missed are all welcome. The version you are looking at fills in automatically.