ChatGPT Prompts for Therapists (Practice Admin, Client Materials, and the Line You Never Cross)

Search "ChatGPT prompts for therapists" and you get two kinds of results, neither of which is written for you. The first is prompts for using ChatGPT as a therapist — the "act as my CBT coach" genre, aimed at consumers. The second is lead magnets from marketing agencies: twenty prompts for Instagram captions, gated behind an email form, sold as a practice-growth course.

The handful written for actual clinicians have a worse problem. They cheerfully suggest you paste session content into a chat window to "generate a SOAP note" or "build a treatment plan based on your client's information." That advice would fail a licensing board review in about ninety seconds. It treats the most sensitive category of record in healthcare as ordinary input text.

There is a real, useful, defensible version of this. It runs along the arc of a private practice — the first email from a stranger, the consult call, intake paperwork, the materials you hand people between sessions, the conversations you dread, and the marketing you do without violating your ethics code. And it starts with one rule that makes everything after it safe.

The rule everything else depends on

Nothing that could identify a client goes into a consumer AI tool. Ever. Not once, not "just this one time," not with the name removed but everything else intact.

If you bill insurance electronically you are almost certainly a HIPAA covered entity, and chatgpt.com is not operating under a business associate agreement with you. Even if you are cash-pay and outside HIPAA's direct reach, your state licensing board and your professional ethics code still hold you to confidentiality. And psychotherapy notes — the process notes you keep separate from the billing record — carry a stricter protection than the rest of the chart, which is exactly why they are the worst possible thing to paste into a chat box.

The workaround is not "be careful." It's a habit: describe the clinical situation in the abstract, and let the specifics live only in your EHR.

Don't type	Type instead
Client names, initials, partner/child names	"a client," "their partner," "a school-age child"
Dates of birth, dates of service, ages under 90	"an adult in their thirties," "early in treatment"
City, employer, school, church, clinic name	"a mid-size employer," "a rural area"
Diagnosis codes attached to a person	the presenting concern in general terms
Verbatim session content or quotes	a two-line summary in your own abstracted words
Anything unusual enough to identify one person	generalize it, or leave it out entirely

That last row is the one people miss. "A client who is a professional cellist recovering from a hand injury after a car accident in a town of 4,000" has no name in it and is completely identifiable. De-identification is about the combination of details, not the name field.

If your EHR offers an AI documentation assistant under a signed BAA, that is a genuinely different legal situation from pasting into a public chat window — evaluate it on its own terms. Nothing below assumes you have one.

The setup prompt (paste once, reuse all year)

You're helping me, a licensed mental health clinician in private practice, with writing and practice administration. My context: I'm a [LPC / LCSW / LMFT / psychologist / LPCC], I practice in [state], my clients are primarily [describe generally: adults with anxiety and burnout, couples, adolescents, trauma-focused work], my modality leans [CBT / ACT / EFT / psychodynamic / IFS / integrative], and my tone is [warm and direct / gentle and spacious / plainspoken].

Absolute rules for this entire conversation: I will never give you protected health information, and you will never ask for it — if a draft would need a client-specific detail, insert [BRACKET] and tell me what to fill in myself. You do not diagnose, do not assess risk, do not produce clinical documentation for a specific person, and do not tell me what a client needs. You are drafting the writing around my clinical work, never the clinical work itself. Where something would need my own clinical judgment, my supervisor or consultation group, my attorney, or my licensing board's rules, flag it as [CLINICAL CALL] or [CHECK BOARD RULES] with one line on why. Everything you produce is a first draft I will edit before anyone sees it.

That block does more work than any individual prompt below. It stops the model from doing the confident, fluent, wrong thing — writing a treatment plan for a person it has never met, in language that sounds authoritative enough that you might not catch it.

1. First contact and the consult call

The inquiry email is the most-repeated writing task in private practice and the one most therapists rewrite from scratch every single time.

Write a reply to a first-contact inquiry from a prospective therapy client. They wrote: [paste their message with all identifying details removed]. Structure: acknowledge that reaching out took something, answer their actual question in one or two sentences, explain how my process works ([free 15-minute consult call], then scheduling), give my fee [BRACKET] and insurance situation [BRACKET] plainly rather than burying it, and offer two concrete next steps. Warm but not effusive, under 180 words. Do not offer clinical impressions or reassurance about whether therapy will help them.

The consult-call question set: "Write 10 questions for a 15-minute consultation call with a prospective client presenting with [general concern]. The goal is mutual fit, not intake. Cover what brought them now, what they've tried, what they're hoping is different in six months, what they need from a therapist, previous therapy experiences that went badly, and logistics. Order them so it feels like a conversation, not a screening. Add one closing question that makes it easy for either of us to say this isn't the right match."

The referral-out message: "Write a kind message to a prospective client I can't take on because [reason: outside my scope, no openings, insurance mismatch, a conflict of interest]. Name the reason honestly without over-explaining, point them toward a concrete next step (a directory, a referral list, their insurer's provider line), and make it clear this isn't a judgment about them. Under 130 words, no apology spiral."

The waitlist email: "Draft a waitlist confirmation that sets honest expectations: roughly how long [BRACKET], that I'll check in [monthly/at X], that they should absolutely keep looking elsewhere in the meantime, and how to reach me if their situation changes. Include a short line of general resources for the interim [I'll fill in local ones] and make it clear this is not clinical support."

The no-show on a consult: "Write a brief, non-shaming follow-up to someone who booked a consult call and didn't attend. One easy path to rebook, one easy path to say they've changed their mind, no guilt, under 80 words."

2. Intake, consent, and policies people actually read

Most practice paperwork is either a lawyer's template nobody finishes or a friendly document that fails to say the hard things clearly. AI is unusually good at the translation problem in between.

Rewrite this practice policy in plain language for a client who is anxious, not a lawyer: [paste your existing policy text — your own, no client information]. Target an eighth-grade reading level, keep every substantive term intact, use short paragraphs and headers, and put the thing people most need to know first. Where the original is ambiguous, don't smooth it over — flag it as [AMBIGUOUS] so I can decide what I actually mean. [CHECK BOARD RULES] on anything where my state or license type may require specific language.

That flag matters. Informed consent, mandated reporting disclosures, telehealth consent, and record-retention language are governed by your state and license type, and a model will produce something that reads beautifully and cites a rule that doesn't apply where you practice. Use it for clarity; verify the substance against your board and, for anything consequential, your attorney or malpractice carrier.

The cancellation policy: "Write a late-cancellation and no-show policy that's firm and unembarrassed. Include the notice window [BRACKET], the fee [BRACKET], what happens in genuine emergencies, and why the policy exists — framed around holding the time, not punishment. Then give me a two-sentence version I can say out loud in a first session without sounding defensive."

The telehealth expectations doc: "Draft a one-page client-facing guide to video sessions: a private space, what happens if the connection drops, that they need to be in [state] during session [CHECK BOARD RULES], the platform, what won't be handled by text or email between sessions, and how crisis situations are handled — with a clear statement that I am not an emergency service and what to do instead. Leave [BRACKET] for my local crisis resources."

The between-sessions communication policy: "Write a short policy on email and portal messaging: what's appropriate between sessions, my typical response window, what will be redirected to session time, and an explicit statement that messages are not monitored for emergencies. Kind but unmistakably clear."

The AI disclosure: "Draft two or three sentences for my informed consent describing how I use AI tools in my practice — administrative drafting and educational materials only, never session content, never clinical decision-making, no client information entered into consumer tools. Written for a client, not a compliance officer." Whether this belongs in your consent form is a judgment call and increasingly a board-level question, but writing it clarifies your own policy either way.

3. Documentation — the only pattern that's actually safe

This is where every other prompt list goes wrong, so let's be precise about the difference.

Unsafe: pasting what happened in session and asking for a progress note. You've just put clinical content about an identifiable person into a system you don't control.

Safe: writing your own note, then asking for help with structure — using generic or already-abstracted language — so you have a reusable format rather than a generated record.

I want a reusable template for [SOAP / DAP / BIRP] progress notes for my practice. Build a blank skeleton with each section labeled, one line explaining what belongs in that section and what doesn't, and three or four bracketed prompts per section to cue me. Use no client information — this is an empty form. Then list the five most common ways clinicians make these notes too long or too speculative, so I can avoid them.

The abstracted-language drill: "Give me 15 examples of behaviorally specific, non-interpretive phrasing for progress notes — the kind of observable language that holds up on review — paired with the vague or conclusory version each one replaces. Generic examples only, no scenario about a real person."

The treatment-plan skeleton: "Build an empty treatment-plan template with sections for presenting concern, client-stated goals in their own words, measurable objectives, interventions, and review dates. Include guidance on what makes an objective measurable versus aspirational. Blank form, no example client — I fill in everything clinical myself."

The policy for your own workflow: "Help me write a one-paragraph internal rule for myself about what I will and won't put into AI tools, so I'm not making the decision case by case at 7pm."

Documentation exists to record your clinical reasoning and to protect you and your client if that record is ever requested. A note generated by a model that inferred plausible-sounding content you never observed is worse than a short note you wrote yourself. The nurses working under the same constraint land in the same place — ChatGPT prompts for nurses walks through the de-identify-first habit applied to charting and handoffs, and the discipline transfers directly.

4. Between-session materials and psychoeducation

Here is where AI genuinely gives you hours back, because it's producing general educational content that never touches a specific person.

Write a one-page client handout explaining [concept: the window of tolerance, thought records, values-based action, the cycle of avoidance, co-regulation, sleep pressure] to someone with no psychology background. Use one concrete everyday analogy, no jargon unless I define it in the same sentence, second person, about 400 words. Include a short "what this looks like in practice" section with three examples and one small thing to try this week. This is general education, not advice for a particular person — no diagnosis, no claims about outcomes.

The three-lengths version: "Give me that as three variants: a full handout, a paragraph I can put in a follow-up message, and two sentences I can say in session. Same substance, different depth."

The worksheet builder: "Design a worksheet for practicing [skill] between sessions. Include a two-sentence rationale, the steps, space prompts for the client to fill in, one worked generic example, and a short troubleshooting section for the three most common ways people get stuck. Keep it to one page and make it usable by someone who's tired."

The reading-level adjustment: "Rewrite that handout for [an adolescent / someone with limited English proficiency / a client who finds worksheets patronizing]. Keep the substance identical. For the adolescent version, drop the therapeutic register entirely."

The group or workshop outline: "Outline a 6-session psychoeducational group on [topic] for [general population]. For each session: the single idea, an opening exercise, the teaching content, a practice activity, and a takeaway. Flag [CLINICAL CALL] where screening, exclusion criteria, or scope questions need my judgment."

Always read these before you hand them to anyone. Models produce confident, slightly-off descriptions of clinical concepts — the shape is right and a detail is wrong — and you are the one whose license is attached to the handout.

5. The conversations you'd rather not have

Draft the email raising my fee. Current [BRACKET], new [BRACKET], effective [BRACKET] with [X] weeks' notice. Requirements: state the change in the first two sentences rather than burying it under gratitude, don't over-justify or apologize, acknowledge that cost is real, name what options exist for anyone for whom this is a barrier [sliding-scale spots / a referral conversation / adjusted frequency], and close warmly. Under 200 words. Give me one version for long-term clients and one for clients who started recently.

Fee increases produce more avoidant writing than almost anything else in practice. If you're also working out what the number should be, the value-framing in how to price your product is written for services generally and transfers cleanly to a caseload — just keep the clinical relationship out of the sales framing.

The termination and transfer letter: "Draft a template for ending treatment when [I'm closing my practice / relocating / the client has met their goals / the work has stalled and a different approach is indicated]. Cover: the reason at an appropriate level of detail, the timeline, what happens to records and how to request them [BRACKET], referral options, and a genuine closing. Respectful, not clinical-sounding. [CHECK BOARD RULES] on notice requirements and abandonment standards in my state."

The repeated-no-show conversation: "Give me a script for raising a pattern of missed sessions with a client — curious rather than punitive, treating the pattern as clinical information worth understanding, while still being clear about the policy. A script for saying it in session, plus a short message if they've stopped responding entirely."

The insurance and superbill explainer: "Write a plain-English explanation of how out-of-network reimbursement works for a client who's never filed a claim: what a superbill is, what they do with it, what questions to ask their insurer, and an honest statement that I can't promise what they'll be reimbursed. Include the exact questions to ask when they call. No guarantees about coverage."

The sliding-scale conversation: "Help me write how I describe my sliding scale — who it's for, how many spots exist, how someone asks without a means test, and how I handle it when spots are full — in a way that's direct and doesn't make anyone perform hardship."

6. Marketing that doesn't violate your ethics code

Therapist marketing has a constraint almost no generic prompt list accounts for: testimonials. Counseling and psychology ethics codes restrict soliciting testimonials from current clients and from anyone vulnerable to undue influence, and the standard marketing playbook — "collect reviews, post client wins" — runs directly into it. Any prompt list that tells you to gather social proof is handing you an ethics complaint.

Rules for all marketing writing in this conversation: no testimonials, no client stories even de-identified, no before-and-after narratives, no claims about outcomes or success rates, no language guaranteeing results, no implied superiority over other clinicians, and no composite "clients" presented as real. My credentials are [BRACKET] and I will not overstate them. Write from the reader's experience, not from my results.

With those rules: write my therapist directory profile. Open with what the reader is experiencing rather than my credentials, describe how I work in concrete terms, say who I'm a good fit for and who I'm not, and close with a low-pressure next step. [X] words, first person, no "safe space," no "journey," no stock phrases every profile already uses.

The website about page: "Write my practice About page from that same profile, longer. Include a short section on what a first session is actually like, since that's what anxious readers want and almost no therapist website says. Same ethics rules."

The niche clarifier: "Give me 5 one-sentence descriptions of who I work best with, each written from the client's presenting experience rather than a diagnostic category, ranked by specificity. No outcome claims."

The blog outline: "Outline a 900-word post on [topic] for people considering therapy. Genuinely useful on its own — not a teaser for booking. Include one myth worth correcting and one thing to try. End with a general, non-urgent line about professional support rather than a hard CTA. No clinical advice for a specific reader."

The social version: "Turn that into a short post: a specific first line that isn't a platitude, the one idea, a concrete example with no real person in it, and a question. Under 180 words, first person." The hook-and-structure patterns in ChatGPT prompts for LinkedIn posts keep this from reading like AI — but override anything there that suggests using client results as proof, because for you it isn't available.

The newsletter rhythm: "Plan a 6-email educational series for people on my list on [theme]. For each: the single idea, the angle, and one action step that isn't 'book a session.'" Newsletter content ideas has formats you can slot in without reinventing a topic every month.

7. The business behind the practice

Help me prepare for consultation group. Here is the clinical question in general terms, with no identifying details: [describe abstractly]. Give me: three ways to frame the question so I get useful input rather than reassurance, the specific things I should ask the group about, and what I might be avoiding in how I'm framing it. Do not answer the clinical question — I'm preparing to ask humans.

The CE and reading digest: "I'm learning [modality/topic]. Build me a 90-day self-study plan: the foundational concepts in order, what to look for in a training, questions to ask a prospective consultant, and how to tell rigorous material from repackaged marketing in this space."

The systems audit: "I spend too long on [task: scheduling, billing follow-up, intake paperwork, note backlog]. Ask me eight diagnostic questions about how I currently do it, then propose three changes ordered by effort-to-relief ratio. Assume solo practice and a limited budget." The operational side of running a one-person business is covered more broadly in the solopreneur small business toolkit.

The email backlog: "Here are the non-clinical admin emails I keep rewriting: [list the situations]. Turn each into a reusable template with [BRACKET] placeholders so I stop composing from scratch."

What never goes into ChatGPT

• Any protected health information. Names, initials, dates, locations, employers, or any combination of details specific enough to identify one person. This includes screenshots, uploaded documents, and "anonymized" vignettes with distinctive facts intact.
• Risk assessment. Suicidality, homicidality, abuse disclosures, and crisis situations are clinical judgment exercised by a licensed human in real time, with your safety protocol and your consultation resources. A model does not know your client, cannot read affect, and will produce fluent, generic, dangerously confident text. Do not use it here at all.
• Diagnosis. Not for a client, not for a "hypothetical" that is obviously your client, not as a second opinion. Diagnosis is a clinical act tied to your license.
• Mandated reporting decisions. When and what to report is determined by your state statute, your board, and often a call to the reporting line or your attorney — not by a chat window, and not on a deadline.
• Subpoenas, records requests, and court involvement. Records requests involving mental health treatment are legally intricate, and psychotherapy notes have their own protections. This is an attorney question, immediately.
• Testimonials, outcome claims, and anything about your effectiveness. Restricted by your ethics code and, in many cases, your state board. A fluent AI draft is exactly the kind of thing that turns into a complaint.
• Clinical supervision. Supervision is a relationship with legal and ethical weight. Use AI to organize what you want to bring; bring it to your supervisor.
• Your own therapy. Worth naming. Models are agreeable by design, and agreeableness is not what the work requires. If you want structured reflection between your own sessions, journaling prompts for anxiety and stress is a better container than a chatbot that will validate whatever you bring it.

Make it repeatable

Save four things: your setup prompt with its rules, your de-identification table, your inquiry and consult-call templates, and your marketing rules block. Keep them in a ChatGPT Project or a plain document you paste from, so every draft starts inside your guardrails instead of outside them.

Then write the hard emails while nothing is hard. The fee increase, the termination letter, the repeated-no-show script, the practice-closure notice — draft them on an ordinary Tuesday and file them. The version you write at 9pm after a difficult session is the version most likely to say something you'd take back, and the whole point of having a draft is that the emotional load is already handled before the moment arrives.

If your practice includes any wellness or coaching work alongside clinical hours, the client-facing structures in ChatGPT prompts for personal trainers cover the same drafting problems under a different scope-of-practice line.

FAQ

Is it a HIPAA violation to use ChatGPT as a therapist? Using AI as a drafting tool isn't inherently a violation — putting protected health information into a service with no business associate agreement is the problem. The distinction is what you type, not that you typed. Every prompt above is built so no PHI ever enters the tool. If you want AI closer to your clinical records, that's a conversation about your EHR vendor and a signed BAA, not about a consumer chat app.

Can ChatGPT write my progress notes? Not from session content, no — that requires putting clinical detail about an identifiable person into a system you don't control, and a generated note can include plausible-sounding content you never observed. Use it to build an empty template and to sharpen your own phrasing habits. You write the note.

What about using it for treatment planning? For a blank structure, useful. For a specific client, no. Treatment planning is clinical reasoning tied to your license and your assessment of a person the model has never met and cannot assess. A model will produce something that looks like a treatment plan with total confidence, which is precisely the risk.

Do I have to tell clients I use AI? It depends on how you're using it and what your board says, and the guidance is moving quickly. Administrative drafting with no client information is a different disclosure question from anything touching session content. Many clinicians are adding a short, plain paragraph to informed consent — the prompt in section 2 helps you write it, but check your board's current guidance and your malpractice carrier before you decide.

Can I use it to respond to a client in crisis? No. Crisis response is real-time clinical judgment with legal consequences, and it belongs to you, your safety protocol, and your consultation resources. Draft your general crisis-resource handout in advance if you like. The moment itself is not a drafting problem.

Will clients be able to tell my emails were AI-drafted? Only if you send the first draft. What people notice is a message that could have been sent to anyone — which is what happens when you skip the setup prompt. Edit for your voice, add the specifics yourself, and cut the one sentence that sounds like a brochure.

How is this different from the usual "20 prompts for therapists" lists? Those are mostly Instagram captions, and the clinical ones tell you to paste client information into a public tool. This set follows the practice in order — inquiry, consult, paperwork, documentation, client materials, hard conversations, marketing, business — and builds the confidentiality and ethics constraints into a setup prompt, so what comes out is a draft you can actually edit and send rather than one you have to throw away.