Beginners in AI

Best AI Prompts for Therapists

James Swierczewski's avatar

James Swierczewski

·

AI summary

Seven AI prompts for licensed mental health professionals that scaffold the documentation and communication work without crossing into clinical decision-making: session notes, treatment plans, insurance justifications, risk documentation, between-session emails, termination prep, and CE reflections. Every output is reviewed by the therapist; AI never provides clinical guidance to a client.

Therapist AI use is uniquely sensitive. The licensing board cares. The client’s privacy is protected by statute. The clinical relationship is the entire point of the work. The seven prompts below take this seriously. None produce clinical content for clients. None replace clinical judgment. They scaffold the documentation and communication work that eats unbilled hours so the therapist can spend their session-prep time on the clinical thinking. This is the therapist slice of the AI Prompt Library, paired with a connector callout for the practice-management tools therapists actually use. For the broader playbook see Claude for Therapists.

Why do most AI therapist-AI workflows produce notes the licensing board would reject and undermine the clinical relationship?

The default therapist-AI risk is letting AI write content that touches clinical work it should not touch: client-facing therapeutic advice, fabricated diagnostic language, between-session interventions delivered through AI. Every prompt below is structured to keep AI in the documentation and communication role and to keep the therapist in the clinical role.

Use AI for the documentation drafting, the insurance write-ups, the CE reflections, the prep work that does not require clinical judgment to land on the client. Always review before any output reaches the client or the chart. Never let AI provide therapeutic content directly to a client. If you do draft any client-facing communication with AI, run it through How to Edit AI Out of Your Writing so it sounds like you, not like a chatbot. When a prompt becomes weekly, graduate it using the Prompt-to-Workflow Ladder.

What are the seven for therapists prompts?

Prompt 1

Session Note Drafter

Most therapists write notes at the end of the day, exhausted. This prompt drafts the structured note from your shorthand so you spend the time on the clinical thinking, not the format.

I just finished a session. Here is my shorthand (de-identified):

CLIENT CONTEXT (no identifying details): [GENERAL: ongoing presentation, modality, session number]
WHAT CAME UP TODAY: [TOPICS, AFFECT, NOTABLE MOMENTS]
WHAT I INTERVENED ON: [INTERVENTIONS USED]
HOMEWORK OR BETWEEN-SESSION WORK ASSIGNED: [IF ANY]
MY CLINICAL THINKING ABOUT THE SESSION: [BRIEF]

Draft a session note in [DAP / SOAP / PIE / your team's format]:

1. DATA / SUBJECTIVE / PROBLEM: client's presentation and what they brought, factually.
2. ASSESSMENT / OBJECTIVE: what I observed: affect, engagement, content themes.
3. PLAN / INTERVENTION: what we did and what comes next.
4. RISK SECTION: any safety considerations to document (suicidality, self-harm, abuse, substance use). If none came up, state that explicitly.
5. PROGRESS toward treatment-plan goals: which goal we touched today and how.

DO NOT invent clinical content. If something is unclear in my shorthand, ask before assuming. Use the language of the modality I am practicing (CBT, IFS, EFT, psychodynamic, etc.). Preserve clinical uncertainty ("appeared," "reported," "endorsed") rather than stating things as fact.

When to use: Within 24 hours of the session, ideally same-day. · Best model: Claude (most disciplined about preserving clinical uncertainty and not fabricating content).

Prompt 2

Treatment Plan Drafter

Treatment plans drafted between sessions are often vague and never revisited. This prompt produces a plan with measurable goals that actually shapes the work.

I am drafting an initial treatment plan for a client (de-identified):

PRESENTING CONCERN: [WHAT BROUGHT THEM IN, IN THEIR WORDS]
MY WORKING CLINICAL FORMULATION: [BRIEF]
MODALITY I USE: [APPROACH]
FREQUENCY AND ESTIMATED DURATION: [WEEKLY / BIWEEKLY, EXPECTED EPISODE LENGTH]
WHAT THEY HAVE NAMED AS THEIR GOAL: [CLIENT GOAL IN THEIR WORDS]

Draft a treatment plan with:

1. PROBLEM STATEMENTS: 2-3 clinically-grounded problem statements derived from the presenting concern, written without pathologizing language.
2. LONG-TERM GOALS: 2-3 goals stated in observable, behavioral, and emotional terms (not "feel better").
3. SHORT-TERM OBJECTIVES: for each long-term goal, 2-3 short-term objectives the client and I can track session-to-session.
4. INTERVENTIONS: the specific techniques or approaches I will use, matched to the modality.
5. SUCCESS INDICATORS: how I will know we are making progress.
6. THE QUESTIONS I should ask the client in our next session to refine this plan WITH them, not for them.

The plan must respect client autonomy. Do not impose pathologizing diagnoses or goals the client did not name.

When to use: After the second or third session, when you have enough clinical picture. · Best model: Claude. The discipline about non-pathologizing language matters.

Prompt 3

Insurance Justification Drafter

Insurance companies want medical-necessity language. Therapists want to write like therapists. This prompt produces the medical-necessity write-up without making you betray your clinical voice.

I need to justify continued sessions for a client (de-identified):

DIAGNOSIS ON FILE: [DSM-5/ICD code if relevant]
SESSIONS COMPLETED: [NUMBER]
CURRENT FUNCTIONAL IMPAIRMENT: [WHERE IT SHOWS UP IN THEIR LIFE]
PROGRESS TO DATE: [WHAT HAS IMPROVED]
REMAINING CLINICAL CONCERNS: [WHAT IS NOT YET RESOLVED]
REASON CONTINUED TREATMENT IS NECESSARY: [CLINICAL REASONING]
EXPECTED ADDITIONAL DURATION: [ESTIMATE]

Draft a continued-treatment justification with:

1. THE MEDICAL-NECESSITY FRAMING: how the impairment fits insurance language without overstating the case.
2. THE PROGRESS REPORTED FACTUALLY: what has changed, in observable terms.
3. THE REMAINING IMPAIRMENT: stated in functional terms (impact on work, relationships, daily life).
4. THE TREATMENT-RESPONSE EVIDENCE: why this client benefits from continued treatment specifically with this modality.
5. THE ALTERNATIVES CONSIDERED: why a lower level of care would not be appropriate at this time.
6. THE EXPECTED COURSE: with a clear endpoint or check-in date.

Use the insurance-acceptable clinical language without sacrificing accuracy. Do not exaggerate impairment to justify continued sessions; do not minimize impairment to seem efficient.

When to use: When insurance requests justification or proactively before authorization expires. · Best model: Claude. Tone discipline matters; insurance reviewers respond to clear, measured, accurate writing.

Prompt 4

Risk-Documentation Quick-Draft

A client raised a safety concern. You documented it in the session, but it needs a formal risk note. This prompt produces it.

A client raised the following risk concern in session (de-identified):

WHAT THEY SAID OR REPORTED: [DIRECT OR PARAPHRASED]
RISK CATEGORY: [SUICIDALITY / SELF-HARM / HARM TO OTHERS / CHILD OR ELDER ABUSE / SUBSTANCE-USE EMERGENCY / OTHER]
WHAT I ASSESSED IN SESSION: [ASSESSMENT QUESTIONS AND RESPONSES]
MY CLINICAL JUDGMENT OF THE RISK LEVEL: [LOW / MODERATE / HIGH, with reasoning]
WHAT WE AGREED ON AS A SAFETY PLAN: [SPECIFIC ELEMENTS]
WHAT FOLLOW-UP I SCHEDULED: [TIMING]
WHO ELSE WAS NOTIFIED OR CONSULTED: [IF APPLICABLE]

Draft a risk documentation note that:

1. STATES THE PRESENTING CONCERN: factually, in clinical language, including the client's own words.
2. DOCUMENTS THE ASSESSMENT: what I assessed and how (instrument used, questions asked).
3. DOCUMENTS THE CLINICAL DECISION: the risk level and my reasoning.
4. DOCUMENTS THE SAFETY PLAN: with the specific elements agreed.
5. DOCUMENTS THE FOLLOW-UP PLAN: timing, escalation criteria.
6. DOCUMENTS THE CONSULTATION OR NOTIFICATION: if applicable.

This is a legal and clinical record. Be precise. Do not invent assessment elements. Do not soften the documentation to make the situation seem less serious than it is.

When to use: Same day as the session, before you leave the office. · Best model: Claude. The discipline about not inventing assessment elements is essential for risk documentation.

Prompt 5

Between-Session Email Drafter

A client emails between sessions with a difficult question. The temptation is to either respond too quickly with reassurance or wait too long. This prompt drafts the measured response.

A client sent this between-session email (de-identified):

[PASTE THE EMAIL or describe the question]

My clinical sense of what they are asking under the surface: [YOUR READ]
My relationship with this client (rapport, prior between-session contact): [BRIEF]
My practice's policy on between-session contact: [BOUNDARIES]
The risk level of the email (if any): [LOW / MODERATE / HIGH]

Draft a response that:

1. ACKNOWLEDGES what they brought, without pathologizing it.
2. HOLDS THE FRAME of session work being the place for the deeper conversation.
3. ADDRESSES any safety considerations if applicable, directly.
4. ANSWERS only what can be answered briefly outside the session container.
5. INVITES them to bring the topic back to our next session.
6. CLOSES warmly, with the next session timing.

Keep under 150 words. Do not therapize over email. Do not give advice that should happen in session. Do not over-soft or under-respond if the email contains a safety element.

When to use: Within 24 hours of receiving the email, but not in the first 30 minutes if you can avoid it. · Best model: Claude. Tone discipline is the entire value.

Prompt 6

Termination Conversation Prep

Termination is the most clinically important conversation of the work and the one most often hurried. This prompt prepares the framing so the ending serves the client.

I am preparing for the termination conversation with a client (de-identified):

LENGTH OF TREATMENT: [DURATION]
ORIGINAL PRESENTING CONCERN: [WHAT BROUGHT THEM IN]
WHAT HAS CHANGED FOR THEM: [PROGRESS]
WHAT IS STILL IN PROGRESS: [WHAT REMAINS]
WHO INITIATED TERMINATION: [CLIENT / THERAPIST / MUTUAL / EXTERNAL: e.g., insurance, move]
MY CLINICAL READINESS for them to end: [READY / BITTERSWEET / CONCERNED]
THE CLIENT'S RELATIONAL PATTERN AROUND ENDINGS: [BRIEF]

Draft a framework for the termination conversation:

1. THE OPENING: how to introduce the topic so it does not blindside them (if therapist-initiated) or feel rushed (if client-initiated).
2. THE REVIEW: a structured look back at the work, anchored to specific moments.
3. THE PROGRESS NAMING: what they did. Not what we did. They did the work.
4. THE RELATIONAL WORK: addressing the ending itself as a relational event, given their pattern around endings.
5. THE WHAT-COMES-NEXT: how they will recognize when to return, where to reach out if needed.
6. THE TERMINATION RITUAL: any closing exercise, letter, or marker appropriate to the modality and the client.
7. WHAT I AM HOLDING for them as they leave: any final clinical reflection to offer.

The conversation honors the relationship. Do not rush it. Do not deny the loss. Do not promise the work was complete if it was not.

When to use: One to two sessions before the planned final session. · Best model: Claude. Tone discipline is essential for endings.

Prompt 7

Continuing Education Reflection

Most CE reflection writing is a chore. This prompt produces a reflection worth re-reading, anchored to your actual practice.

I just completed [CE COURSE / WORKSHOP / TRAINING ON TOPIC].

Key concepts I learned: [BULLET POINTS]
What surprised me or challenged my prior thinking: [BRIEF]
A recent case (de-identified) where this would have changed my approach: [BRIEF]
What I want to integrate into my practice going forward: [SPECIFIC]

Draft a CE reflection (500-700 words) with:

1. OPENING: the question or clinical situation that drew me to this CE.
2. KEY LEARNINGS: 2-3 specific concepts, each tied to my actual practice (not generic restatement of the curriculum).
3. THE CHALLENGE TO MY PRIOR THINKING: where this course pushed back on something I assumed.
4. THE CASE APPLICATION: the de-identified clinical situation where I would have done something different had I known this.
5. THE INTEGRATION PLAN: 2-3 specific changes I will make to my work, with how I will know they took root.
6. THE NEXT QUESTION this course raised that I still want to explore.

Do not write generic praise of the course. Be specific. Reflections that read as performative get flagged by licensing boards; reflections that read as actual integration earn the CE credit and improve the work.

When to use: Within a week of the course completion. · Best model: Claude. Tone discipline avoids the performative-reflection trap.

These work across Claude, ChatGPT, Gemini, and Grok. Claude is the strongest default for mental health work because of its discipline about preserving clinical uncertainty (“appeared,” “reported,” “endorsed”) and not pathologizing client language. For HIPAA compliance and your state licensing board’s view on AI, the tool choice matters: use only paid plans with reviewed data-handling terms and a signed Business Associate Agreement where PHI is involved. De-identify all inputs by default; the prompts above are written to use general descriptors instead of identifying detail.

What is the worst thing you can do with AI for therapists?

Three patterns will burn therapists fastest, and all three have licensing weight.

  • Pasting identifiable client information into a free-tier AI tool. Free tiers may train on inputs and may retain content. This is a HIPAA violation and likely a violation of your state licensing board’s ethics rules. Use paid plans with verified data-handling and, where required, a Business Associate Agreement.
  • Letting AI produce content that goes to the client without your clinical review. AI generates plausible-sounding therapeutic content that can be wrong in clinically significant ways. Every patient-facing or client-facing output must be reviewed by the licensed therapist before transmission. AI is the documentation co-pilot, never the clinician.
  • Asking AI for specific diagnostic criteria, DSM codes, or modality-specific intervention scripts as if they were authoritative. AI fabricates clinical detail confidently. Use AI to structure your reasoning; reference authoritative clinical sources (DSM-5-TR, ICD-11, your modality’s manual, peer-reviewed literature) for the substance.

What if you want to take this further?

Each prompt above takes inputs you paste in. The next move is connecting AI to the practice-management systems where therapist work already lives, with extra care for PHI handling.

Connectors are now standard

Claude, ChatGPT, and Grok all support connectors that let your AI read live data from your work tools (Gmail, Notion, GitHub, Asana, HubSpot, Stripe, and many more) instead of relying on you to paste context. For therapists this means the AI can read appropriate non-PHI workflow tools (your calendar, your scheduling system, your CE tracker) while staying out of direct EHR access where institutional policy and HIPAA preclude it.

For therapists, the connectors worth pairing with these prompts:

  • Calendar connector — for session-prep workflows that need to reference appointment timing without pulling in chart content.
  • Notion connector — if you maintain non-PHI clinical templates, CE tracking, or practice ops in Notion, AI reads them for consistency.
  • Gmail or Outlook connector — for the between-session email and insurance-justification prompts; AI references prior correspondence WITHOUT pulling in protected health information from secure messaging portals.
  • Google Drive connector — for de-identified case discussion, peer consultation prep, or research summary work.
  • DO NOT use — consumer AI connectors with your actual EHR (SimplePractice, TheraNest, Headway, Alma) unless your platform has explicitly approved the integration with a signed BAA and audited data flow.

What are common questions about AI for therapists?

Is AI use in clinical practice ethical for therapists?

It depends on the tool, the plan, the data, and your licensing board’s current position. Many state boards now have specific AI guidance. The line that holds across most boards: AI may scaffold documentation, drafting, and administrative work, but cannot substitute for clinical judgment and cannot deliver therapy to clients. Check your specific state board’s published position before integrating any AI tool into client-facing work.

Will AI replace therapists?

AI is changing what therapists do, not eliminating the role. Documentation, insurance work, CE reflection, scheduling: all compressing. The therapeutic relationship, the moment-to-moment attunement, the read of the client in the room: not. Therapists who use AI for the back-office work and spend their saved time on actual clinical thinking become better, not worse. Therapists who use AI to deliver therapy create harm.

Which AI tool is best for therapists?

For documentation drafting on de-identified content, Claude Pro is most disciplined about preserving clinical uncertainty. For PHI handling, you need a HIPAA-compliant tool with a BAA: Claude Enterprise, ChatGPT Enterprise with BAA, or therapy-specific AI tools (Mentalyc, Upheal, Eleos Health). Verify what your practice or platform has approved. Do not use unapproved tools for clinical work.

Can AI write my session notes?

Some therapy-specific AI tools generate notes from session audio. Whether this is appropriate depends on client consent, your state board’s view, and your platform’s policy. General-purpose AI is appropriate for editing and structuring notes you outline yourself; it should not generate clinical content from scratch. Always review and sign every note.

Should I tell my clients I use AI?

Increasingly required by some state ethics rules. The line: if AI affects a clinical decision or generates client-facing content, disclose. If AI assists with documentation or workflow only, most boards do not require disclosure but transparency is increasingly expected. Check your state board’s position. Discuss with your client at the start of treatment as part of informed consent if AI use will touch their record.

How do I avoid AI fabrication in clinical notes?

Never ask AI to fill in clinical detail you did not provide. Use AI to structure what YOU observed and reasoned; never let AI generate clinical content from a blank prompt. Every prompt above is built to keep AI in the structural role and to require therapist review. If AI generates a clinical detail you did not provide as input, do not use it.

How long does it take to build the therapist-AI loop?

Six weeks. Start with the Session Note Drafter on de-identified inputs first to learn the rhythm. Add the Insurance Justification prompt the next time you need to write one. Add the Risk Documentation prompt only after you have established your rhythm with the others. Most therapists settle into 3-4 of the seven prompts within a quarter.

🎯

The AI Prompt Library · $39

Clinical practice workflows, prompt-paved.

Soon to be 1000+ prompts in Notion organized by use case. The full therapist section includes everything above plus prompts for intake assessment drafting, group therapy prep, supervisor consultation notes, practice marketing, and self-of-the-therapist reflection. Plus prompts for every other field. Lifetime access.

Get the Library →

Get Smarter About AI Every Morning

Free daily newsletter. Built for people who want to use AI well, not chase every model.

Free forever. Unsubscribe anytime.

Sources to read next?

You might also like

Two ways to go further

The AI Prompt Library

1,000+ ready-to-use prompts for Claude, ChatGPT, and Gemini. Stop staring at a blank box.

Get it for $39 →

2-Hour Live AI Crash Course

A private, beginner-friendly session across Claude, ChatGPT, Gemini, and the wider landscape.

Book for $125 →

Discover more from Beginners in AI

Subscribe now to keep reading and get access to the full archive.

Continue reading