AI Prompts for Coaches and Therapists (Practice Admin)

TL;DR: AI can genuinely help a coaching or therapy practice with website copy, scheduling and cancellation emails, intake logistics, invoicing reminders, workshop marketing, newsletters, and draft business policies. It should never touch session notes, treatment plans, assessments, diagnoses, or anything a client told you. Every prompt below is scoped to the business, not the clinical work, and the boundary comes before the first one.

Running a coaching or therapy practice is two jobs wearing one name badge. There is the work with clients, which is the actual craft, and there is everything that makes the work possible: a website that explains what you do, an email confirming Tuesday's appointment, an invoice that needs paying, a policy that says what happens when someone no-shows. The second job has no clinical content in it at all, and it is also the job that eats the evening you'd rather spend on the first one.

That second job is what this article is for. Every prompt below produces something you would be comfortable showing to anyone — a colleague, a landlord, a client who never worked with you — because none of it describes a real client's situation. If a prompt or a draft ever starts to include something a client told you, stop, and finish that piece of writing yourself.

The instinct to keep the two jobs separate is usually right, but "usually right" is not the same as knowing exactly where the edge sits, and most practitioners have never had to draw that line explicitly before an AI tool made it a live question. Nobody trained you on which parts of running a practice are safe to hand to a language model, because the tools are new enough that the training barely exists yet in most professional programs. So the line drawn below is deliberately conservative: where a use case could plausibly drift toward clinical content, it is marked as needing extra care, or excluded outright, rather than left to individual judgment on a busy Tuesday.

Where exactly is the line between practice admin and clinical work?

Admin work is anything that stays true regardless of which client you're thinking about. Clinical work is anything that only exists because of one specific client's specific situation.

A quick test for anything not on this list: would the exact same sentence be true and appropriate for every single client on your calendar, or does it only make sense because of one particular person's situation. A cancellation policy is true for everyone. A workshop start date is true for everyone who signs up. A note about what someone said in a session is true for exactly one person, which is the signal that it belongs in a record, not a prompt. When a piece of writing fails that test, the fix is almost always to write the general version yourself first and let a model help only with the parts that stay general.

Why can't AI tools touch clinical content at all?

Because what a general-purpose AI vendor does with what you type depends on the product, the account type, and a setting you may not have checked — and none of that is something a feature list settles.

Read the vendors' own words rather than trust a marketing page. Google's Gemini Apps Privacy Hub states plainly: "Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services, including machine-learning technologies." The same page adds that some reviewed chats are not deleted when you delete your activity — "Chats reviewed by human reviewers (and related data like your language, device type, location info, or feedback) are not deleted when you delete your activity. Instead, they are retained for up to three years."

Anthropic's Claude Privacy Center draws a specific line by product tier: "This article is about our consumer products such as Claude Free, Pro, Max and when accounts from those plans use Claude Code." Consumer conversations are not used to train models by default, only if you opt in, if a conversation is flagged for safety review, or through an explicit program like a trusted-tester scheme — a materially different default than "everything you type may be reviewed", but still not something to bet client confidentiality on without checking your own account's settings.

OpenAI's Data Controls FAQ describes a similar per-account choice: "Data Controls let you decide how ChatGPT uses your conversations and interactions." Even its most privacy-protective mode, Temporary Chat, is not a vault — OpenAI states that "Temporary Chats are deleted from our systems after 30 days", which also means they exist, readable, for those 30 days.

None of that is a reason to relax the rule above. It is the reason for it. Three different vendors, three different defaults, and every one of them requires you to check a setting rather than assume a protection exists. The rule that survives all three, that nothing clinical goes in regardless of settings, is simpler and safer than tracking which vendor does what this month.

What about AI features already built into my scheduling or EHR software?

A growing number of practice-management and scheduling platforms now advertise built-in AI features: automated appointment summaries, AI-drafted intake follow-ups, or a chat assistant layered onto a system you already use. The caution from the section above applies to those too, and arguably applies harder, because a practice-management vendor's marketing copy is more likely to gesture at compliance than a general-purpose chatbot's is.

A feature list is not a signed agreement, and a vendor's homepage claiming a product is "built for healthcare" or "practice-ready" is not the same as that vendor agreeing, in writing, to handle the specific category of data your work involves. Before turning on any AI feature inside software that already touches client records, ask the vendor directly three things: what data the feature sends outside the system it currently lives in, whether that transfer is covered by whatever agreement you already have with them, and whether the feature can be scoped to admin fields only rather than clinical notes. If a vendor cannot answer those clearly, treat the feature as unverified rather than assuming the rest of the product's safeguards extend to it automatically.

This is the same rule as the rest of this article, applied to a tool you may already be paying for rather than one you are considering adding. New AI features get bolted onto existing practice software constantly, and each one is a fresh question, not an extension of trust you already gave the base product.

What can AI actually help a coaching or therapy practice with?

Everything below produces a draft for you to read, correct, and personalise before it goes anywhere near a client. None of it needs a real client's name, situation, or history to work.

Website and service-page copy

Write a 150-word "About" section for my [coaching/therapy] practice website. My focus areas
are [list, e.g., career transitions, relationship coaching]. My credentials are exactly:
[list them precisely — do not add, infer, or upgrade any credential]. Tone: warm, professional,
plain language, no jargon. Make no claims about outcomes, cure rates, or guarantees. End with
one sentence inviting a visitor to book a consultation call.

Draft a service-page description for a [session type, e.g., "60-minute individual coaching
session"] priced at $[amount]. Describe what a first session logistically involves — format,
length, what to expect administratively — without describing specific techniques as guarantees
or promising any result. Add a 3-question FAQ covering only logistics: scheduling, cancellation,
and payment.

Scheduling and cancellation emails

Write a friendly email confirming an upcoming appointment. Include: the client's first name
only [I will insert], the date and time [I will insert], the format (in-person or telehealth),
and one line restating my cancellation policy, which is exactly: [paste your actual policy
text]. Do not add anything about what we will discuss in the session.

Write a short, polite email for when a client misses a session without notice. State that the
missed-session fee is $[amount] per my cancellation policy, offer to reschedule, and ask them
to confirm they received the note. Keep the tone non-judgmental and do not speculate about why
they missed it.

Intake logistics, never intake content

Write a pre-first-session logistics email for a new client. Cover: which forms to complete
before the appointment, named only by title (e.g., "New Client Intake Form", "Cancellation
Policy Acknowledgment") — never the questions inside them — where to park or how to join the
telehealth link, which payment methods I accept, and how long the first session runs.

Invoicing and payment reminders

Write a polite payment reminder email for an invoice that is 7 days overdue. Invoice number
[insert], amount $[insert], original due date [insert]. Offer the payment methods I accept:
[insert]. Keep it to one short paragraph, neutral in tone, not accusatory.

Workshop and group-program marketing

Write marketing copy for a 4-week [topic] group workshop, [format], starting [date], capped at
[number] participants, priced at $[amount] per person. List three concrete logistical
takeaways a participant leaves each session with, described in general terms (for example,
"a written weekly plan template") rather than as a clinical or therapeutic promise. End with
a short call to action to join the waitlist.

Newsletters to your mailing list

Write a monthly newsletter for my practice's mailing list with three sections: one practice
update (I will supply the specific news — new hours, a workshop, a policy change), one general,
non-individualised tip that is educational rather than directed at any specific person's
situation, and one short call to action to book a session. Keep the general tip broad enough
that it could not read as advice to a particular reader's circumstances.

Business policies, drafted for your own review

Draft a plain-language cancellation and no-show policy for my practice. Cover: how much notice
is required, the fee for a late cancellation or no-show, any exceptions I want to allow
[insert], and how a client can cancel or reschedule. Write it explicitly as a draft for me to
check against my own professional body's guidance and any legal requirement where I practise —
do not present it as a finished, legally reviewed policy.

A referral thank-you and a review request

Write a short thank-you note to a colleague who referred a new client to my practice. Do not
mention the client's name or any detail about them. Keep it warm and brief, one paragraph.

Write a short email asking a long-term client if they would be willing to leave a general
Google review about their experience with my practice, without describing what we worked on
or referencing anything from a session. State plainly that the review should stay focused on
logistics and overall experience — scheduling, communication, professionalism — not treatment
content, and that leaving one is entirely optional.

What should never go into a prompt, even by accident?

The most common failure isn't deciding to paste clinical content on purpose. It's personalising a safe template and typing a real detail into it without noticing, writing "confirming Tuesday's session, following up on what you mentioned about your manager" instead of the neutral logistics line the template actually needed.

A safe rewrite of that same reminder keeps every operational detail and drops the rest: "Confirming your appointment for Tuesday at 2pm, telehealth link attached below. Let us know if you need to reschedule." Nothing in that sentence would change if it were being sent to a different client entirely, which is exactly the test worth applying before any admin email goes out. Read the draft back and ask whether a stranger who intercepted it would learn anything about why the recipient is a client at all. If the answer is yes, something in the sentence needs to come out.

Build the habit of writing every admin prompt as if a stranger will read it, because with most consumer AI tools, per the vendors' own pages above, one might. If a sentence would only make sense to someone who knows why a specific client is in your care, it does not belong in the prompt. Rewrite it so the sentence would be true for any client on your calendar, or leave it out and add the personal detail yourself, by hand, after the draft comes back.

Does using AI here require a HIPAA- or GDPR-compliant tool?

No single answer, and no plan or feature list settles it on its own. Compliance with a health-privacy framework, wherever one applies to your work, comes from a signed agreement between you and the vendor covering the specific data involved and how the product is configured — not from a marketing page listing what a plan includes. Nothing in this article, and no claim from Prompt Architects or any other AI product, should be read as saying a tool is safe for protected health information without that agreement in place.

Confidentiality, licensing, and record-keeping obligations vary by profession, credential, and jurisdiction. Coaching is regulated loosely or not at all in many places, while licensed therapy typically sits under a specific board or licensing body; either way, the accurate rule for your situation is the one your own regulator or professional body publishes, not a general description in a blog post. If any part of your work is covered by a legal health-privacy framework, that is a conversation to have with a compliance advisor and the vendor directly, before you use a consumer AI tool for anything connected to it.

How do you keep AI drafts from sounding like a form letter?

Give the model your actual wording instead of a description of your tone. Paste two or three things you have genuinely sent — an email, a paragraph from your site, a line from a policy — and ask it to match that voice rather than inventing one. A saved personal context with your practice's name, focus areas, standard policies, and typical phrasing beats retyping the same description every time you open a new chat.

The other habit worth keeping: read every draft this article helps you produce before it goes out, the same way you would read anything with your name on it. A model can produce a fluent invoice reminder or a workshop description in seconds; it cannot know your actual fee schedule, your actual policy wording, or what you actually meant to say to a specific colleague. Verify names, dates, amounts, and any factual claim about your services every time.

Building a small library of your own admin templates, once, pays off every week after. A saved cancellation-policy line, a standard invoicing paragraph, a workshop-announcement skeleton: each only needs writing carefully a single time, and after that the work is filling in a specific date, amount, or workshop title rather than starting from a blank page. That is a smaller version of the same discipline professional writers and marketers already use, adapted to a practice that has no marketing department and one person doing every job in it.

For the marketing side specifically, our landing-page copy prompts and client-update email templates both work well adapted to a practice, since neither assumes anything about the reader's personal situation. If you run any kind of organisation where the same admin-versus-mission-critical-work boundary applies, our piece on AI for churches and faith organisations draws the same line for a different kind of practice.