How to Use Claude as a Therapist's Documentation Aid: The Sophia Skill Guide

Updated

If you came here looking for support for yourself, read this part first. This page is a documentation guide for qualified therapists, and the product on it is a note-writing tool, not therapy. But the search that brought you here is a real one and it deserves a real answer, so the first section below is written for you rather than for clinicians. If you are thinking about harming yourself, please do not work that out alone in a chat window: contact your local emergency services or a crisis line in your country today.

If you are the one looking for support

People search this for good reasons. Therapy is expensive, waiting lists are long, three in the morning is not a time when anyone is available, and sometimes you are not ready to say a thing out loud to a person yet. None of that is silly and none of it deserves a lecture.

0:15 · loop Same AI, same question. One file makes it a specialist.

So, honestly: a chat window is not nothing. It is a place to write, at any hour, without managing anyone else's reaction. It can help you put words to something before you have to say it. It is good at drafting what you want to tell a GP when you have ten minutes and go blank. It can explain a term someone used, or help you think through a decision by asking what you have not considered.

What it cannot do is the part that matters most, and the reason is specific. It agrees with the account you bring. If you describe a situation in the shape your distress has given it, you get back a thoughtful, well-written reply built on that shape, and it will sound like understanding. A therapist notices what you did not say, hears the thing you skated past, and is willing to disagree with you about your own story. A chat only has the words you typed. It also has no continuity: it cannot see you getting worse over six weeks, it is not accountable to anyone, and it cannot do anything.

On cost and access, the practical routes people forget: a GP or family doctor can refer, and in many countries that is the cheapest door. Employers often have an assistance programme nobody uses. Universities and colleges have free counselling for students. Training clinics, charities and professional bodies frequently run low-cost or sliding-scale services, and they are usually not advertised where you would look. Waiting for one of those while writing things down in the meantime is a reasonable plan; using the chat as the plan is not.

And plainly, once: if you are thinking about harming yourself, contact your local emergency services or a crisis line in your country today. That is a conversation for a person, and it is worth having sooner than it feels necessary. The rest of this page is written for therapists doing their paperwork; the AI health prompts guide covers personal use more fully, including what is safe to paste anywhere.

For qualified clinicians

The rest of this page is for registered therapists and counsellors. The skill behind it is Sophia - Therapist AI Skill: a documentation aid that drafts modality-appropriate session notes and treatment-plan frameworks for you to correct and sign - $19, one payment, yours permanently. It does not provide therapy, and it never completes a risk assessment.

View the Sophia skill →

The paperwork problem, stated accurately

Session notes, treatment plans, letters, intake structures and practice admin take up a real share of every client hour, and unlike the clinical work they do not get easier with experience. They also arrive at the worst time: after the session, when the energy has gone. That is the load a documentation aid addresses, and it is the only load it addresses.

The distinction that governs everything below: the model handles structure and prose. You supply every piece of clinical content, and you sign the record. A session note is a legal clinical document. Nothing here changes who is responsible for it.

What is safe to delegate, and what is not

Task Verdict Why
Turning your own post-session account into a structured note in SOAP, DAP or BIRP Genuinely useful It is reformatting what you already decided, which is where the time goes
Treatment-plan and intake frameworks to adapt Useful as a skeleton The structure is generic; the content is yours and must be
Practice documents: contracts, privacy notices, cancellation policies Useful, then reviewed Standard documents, but they carry legal weight in your jurisdiction
End-of-therapy and referral letters Useful Register and structure, from content you provide
Inferring what the client meant, or what is going on clinically Never It was not in the room and has only your words. A fluent inference is still an invention
Risk assessment, in any part Never This is clinical judgement with consequences. A framework to complete yourself is the furthest this goes
Diagnosis, formulation, or a decision about a client's care Never Yours, professionally and legally
Anything pasted with identifying details before you have checked your obligations Settle this first Confidentiality comes before any time saving. See below

Before you paste anything: the confidentiality question

Decide this once, properly, rather than per session. Your professional body's ethical framework, your information-governance obligations and, in the UK and EU, GDPR treatment of special category data all bear on what may go into a third-party tool, and the answer depends on your tier, your agreements and your jurisdiction rather than on the tool being useful. Practices that work regardless: draft from [CLIENT] placeholders rather than names, leave out dates of birth, addresses and identifying circumstances, and remember that a rare presentation plus a small town can identify someone as surely as a name can. If you are employed, this is a question for your information-governance lead before it is a question for you.

Prompt 1 - the session note that adds nothing

The whole value is in what this forbids. A note that reads better than your recollection is a liability, not a time saving.

Draft a session note in [SOAP / DAP / BIRP / my practice format] for a session conducted in [MODALITY]. Below is my own account of the session, written from my notes. Refer to the client as [CLIENT] throughout. Hard rules, which override anything I ask later: - Use only what I have written. Do not infer content, meaning, affect, motivation or clinical significance that I did not state. - If a required field of this format has nothing in my account, write [NOT RECORDED] rather than filling it. - Do not add clinical language I did not use. Do not upgrade "seemed tired" into anything diagnostic. - Quote the client only where I have quoted them. - Do not draft, complete or suggest any part of a risk assessment. If my account touches on risk, write [RISK: CLINICIAN TO COMPLETE] and say nothing further. - Keep it to the length the content justifies. After the note, list separately: anything in my account that was ambiguous and how you resolved it, and every field you marked [NOT RECORDED]. MY ACCOUNT: [PASTE]
Skill spotlight
Sophia - Therapist AI Skill
Sophia - Therapist AI Skill
$19 · one payment

Holds your modalities, your note format and your practice house style between sessions, so you are not respecifying the structure every evening. Built to draft only from what you record, to mark what you did not state rather than filling it in, and to stop at anything requiring clinical judgement. A documentation aid for qualified clinicians; it does not provide therapy.

View Sophia - Therapist AI Skill →

Prompt 2 - a treatment-plan framework with the clinical content left blank

Give me a treatment-plan framework for work in [MODALITY] with a presenting issue of [GENERAL DESCRIPTION, NO IDENTIFYING DETAIL]. Produce the structure only: - the sections this plan should contain, in order, for this modality - for each section, one line on what belongs there and what does not - every clinical field left as [CLINICIAN], with no suggested content, no example goals and no sample interventions - the review points a plan of this kind usually specifies, as blanks Then list the questions I should have answered with the client before this plan can be completed, and the elements my professional body or service is likely to require that a generic framework would omit. Suggest nothing clinical. If you are tempted to propose a goal or an intervention, list the question I should ask instead.

Prompt 3 - the identifier sweep, before anything leaves your notes

This checks text you are about to reuse. It is a second pair of eyes, not a clearance: the judgement stays yours.

Below is text I am preparing to use outside my clinical record. Do not rewrite it and do not summarise it. List every element in it that could identify a person, directly or in combination, including: - names, initials, relationships and job titles - dates, ages and time references that fix a period - places, services, employers and institutions - unusual circumstances, conditions or events that would be recognisable in a small population - quoted phrasing distinctive enough to be traced For each, quote it and say whether it identifies alone or only in combination with something else in the text. Then state plainly what would still be identifiable if every name were removed. Do not tell me the text is safe; that is my judgement and my professional obligation, not yours. TEXT: [PASTE]

Prompt 4 - the end-of-therapy or referral letter

Draft an [ending letter / referral letter / summary for onward care] from the content below. Recipient: [WHO]. Refer to the client as [CLIENT]. Rules: - Content comes only from what I provide. Add no history, no formulation, no outcome and no recommendation I did not write. - Match the register to the recipient: a GP, another clinician and the client themselves need three different letters. Ask me which if I have not said. - Where a fact is needed and missing, write [CLINICIAN TO ADD]. - No risk content unless I have written it, and then reproduce it exactly as I wrote it rather than rephrasing. - Keep it short. Length is not thoroughness. Then list anything in my source material that I may have intended for my own record rather than for this recipient.
Complementary prompt pack
Healthcare Documentation Prompt Library
Healthcare Documentation Prompt Library
$9 · one payment

The wider clinical correspondence around a caseload - clinic letters, case summaries, referral and discharge writing, patient-facing explanation - as ready prompts rather than a persona you load. Same rule throughout: drafts for the clinician to review and sign.

View Healthcare Documentation Prompt Library →

Where it fails

  • Fluency that outruns your recollection. The characteristic failure is not a wrong note, it is a note that reads more confidently than you actually remember the session. Anything you did not state belongs in square brackets, not in prose.
  • Quiet clinical upgrades. "Seemed flat" becomes something diagnostic; a passing remark becomes a theme. Instruct against it explicitly and check for it anyway.
  • Risk. It will draft a risk assessment if you let it. Do not let it. A blank framework is the limit.
  • Agreeableness. Ask whether a note is adequate and you will be told yes. Ask what a supervisor or an auditor would query in it.
  • Jurisdiction. Record-keeping requirements, retention periods and what a letter may contain differ by country, professional body and service. It blends them.
  • Confidentiality at the moment of pasting. There is no later fix, which is why that decision belongs before the workflow rather than inside it.
  • It is not supervision. Nothing here is a substitute for clinical supervision, and using it as a place to think through a difficult case is a different activity with different confidentiality implications.

How do you install the Sophia skill?

The download is a ZIP with SKILL.md at the root of the archive, not inside a nested folder - that folder structure is the usual reason an upload fails. In the Claude desktop app, open Customize → Skills, upload the ZIP and toggle it on.

Skills need code execution enabled, under Settings → Capabilities. Anthropic's help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise - so if you are on the free plan, check Settings → Capabilities for your own account rather than taking either page's word for it. On a service-managed account that switch sits with an owner, and in a healthcare organisation it may have been turned off deliberately - which is worth asking about rather than routing around. The full walkthrough is in the skill installation guide.

In ChatGPT or Gemini there is no upload step: open SKILL.md, copy the contents, and paste them into custom instructions. You lose automatic triggering and keep the method.

Who is this for?

Qualified, registered therapists and counsellors in private practice or a service, carrying a caseload and its paperwork. It works in Claude, ChatGPT or any AI chat. The adjacent clinical roles are kept separate in the catalogue rather than merged into one vague healthcare assistant:

The wider set is the healthcare, science & education skills, and the same documentation discipline from the regulatory side is in the medical writer guide.

In summary:

For clinicians: the model handles structure and prose, you supply every piece of clinical content, and you sign the record. Draft only from what you wrote, force [NOT RECORDED] rather than inference, never let it touch a risk assessment, and settle the confidentiality question before the first paste rather than during it. For the note-writing itself, Sophia - Therapist AI Skill ($19); for the wider clinical correspondence, the Healthcare Documentation Prompt Library ($9). Both work in Claude, ChatGPT and any AI chat, with a 30-day money-back guarantee. A documentation aid only, and not therapy, supervision or clinical advice.

Skill · .md · For qualified clinicians

Sophia - Therapist AI Skill

One file that keeps your modality, note format and house style between sessions, drafts only from what you recorded, marks what you did not state instead of inventing it, and stops at anything that needs clinical judgement. No subscription. Yours permanently.

$19
Get the Sophia skill →

KissMySkills is a marketplace of 2,000+ AI skills, 300+ prompt packs, 100+ agents & free tools for Claude, ChatGPT & any AI chat.

Related Skill Guides

Put this into practice

Or get every skill with Library See plans →

Frequently asked questions

Can I use Claude as a therapist for myself?+

No, and it is worth being specific about why rather than just saying no. A chat window is a place to write at any hour, and it can help you put words to something, draft what you want to tell a GP, or explain a term. What it cannot do is the part that matters: it agrees with the account you bring, so if you describe a situation in the shape your distress has given it, you get a thoughtful reply built on that shape that sounds like understanding. A therapist notices what you did not say and is willing to disagree with you about your own story. It also has no continuity, cannot see you getting worse over weeks, is not accountable, and cannot act. If you are thinking about harming yourself, contact your local emergency services or a crisis line in your country today.

What is this skill actually for?+

Documentation, for qualified and registered therapists and counsellors. It drafts modality-appropriate session notes in formats such as SOAP, DAP and BIRP from your own post-session account, plus treatment-plan frameworks, intake structures, ending and referral letters, and practice documents. It does not provide therapy, counselling or mental-health advice, and every document it produces is a draft for the clinician to correct, personalise and sign.

Can AI write my session notes?+

It can reformat your own account into your note format, which is where the time actually goes. It must not add anything. The instruction that makes this safe is to forbid inference: use only what the clinician wrote, mark any required field with nothing behind it as NOT RECORDED rather than filling it, and never upgrade an observation such as seemed tired into clinical language. A note that reads more confidently than you remember the session is a liability rather than a time saving.

Can AI complete a risk assessment?+

No, and this is the hardest line on the page. Risk assessment is clinical judgement with consequences, and a model will draft one fluently if you let it. A blank framework for the clinician to complete is the furthest this goes. A useful instruction in any note prompt is that if the account touches on risk, the model writes RISK: CLINICIAN TO COMPLETE and says nothing further.

Is it safe to put client information into an AI chat?+

Settle this once, before the workflow rather than inside it. Your professional body's ethical framework, your information-governance obligations and, in the UK and EU, GDPR treatment of special category data all bear on it, and the answer depends on your tier, your agreements and your jurisdiction. Practices that work regardless: draft from CLIENT placeholders, leave out dates of birth, addresses and identifying circumstances, and remember that a rare presentation plus a small town identifies someone as surely as a name does. If you are employed, this is a question for your information-governance lead first.

Who is responsible for a note drafted by AI?+

The clinician, without qualification. A session note is a legal clinical record. The model handles structure and prose; the clinician supplies every piece of clinical content, checks the draft against what actually happened, and signs it. Nothing about using a drafting tool changes professional or legal responsibility for the record.

What does it get wrong most often?+

Fluency that outruns your recollection, and quiet clinical upgrades where an observation becomes a diagnosis or a passing remark becomes a theme. It is also agreeable: ask whether a note is adequate and you will be told yes, so ask instead what a supervisor or an auditor would query. And it blends jurisdictions, since record-keeping requirements, retention periods and what a letter may contain differ by country, professional body and service.

Is this a substitute for clinical supervision?+

No. Nothing here replaces supervision, and using a chat as a place to think through a difficult case is a different activity from drafting a note, with different confidentiality implications that should be considered separately.

~/get-started

Skills that work. No fluff.

Browse every skill, prompt pack, and agent in the store.

Browse all skills →Or try the free tools