Use AI for the paperwork around a group, not the clinical decisions. It can draft a session-by-session programme outline, a screening questionnaire, invitation and waitlist emails, and a fee-and-capacity model in an afternoon. You decide who is suitable, what the group promises and what gets sent. Keep identifiable enquirer details out of general chat tools.
Most groups that never start fail in the same place: not the idea, but the gap between a good outline and six people booked and paid. The workflow below is built around that gap, in the order you meet it: numbers first, then the outline, then screening, then the invitation, then the waitlist.
Work out whether the group can run before you write anything
An outline written before you have tested the numbers often has to be rewritten. Start with a short conversation with an AI assistant that holds no client information at all. You are only giving it figures about your own practice.
Say a solo therapist with a private room is planning an eight-week group. These are illustrative figures, not benchmarks:
| Line | Figure | Note |
|---|---|---|
| Fee per member per session | $45 | Your own pricing decision |
| Sessions | 8 at 90 minutes | Plus 30 minutes of notes and admin each week |
| Individual rate for comparison | $110 per 50 minutes | Your normal fee |
| Room hire | $25 per session | Zero if you use your own room |
| Pre-group screening calls | 10 at 20 minutes | Unpaid unless you charge an assessment fee |
| Supervision allowance | 2 hours at $90 | Optional, but sensible for a first group |
Revenue at six members is 6 x $45 x 8 = $2,160. At eight members it is $2,880. At four members it is $1,440. Now the time: eight sessions at two hours each (90 minutes plus 30 of notes) is 16 hours, plus about 3.3 hours of screening calls, plus two supervision hours, plus roughly 3 hours of preparation, which comes to about 24 hours. Room hire adds $200 and supervision $180, so costs outside your time are $380.
At six members, income after those costs is $2,160 minus $380 = $1,780, or about $74 an hour for the 24 hours. At four members it is $1,060, or about $44 an hour. Compare that with $110 for a 50-minute individual session, which is about $132 an hour, and you can see whether the group makes sense for you. It often does not pay more per hour, but it can serve more people and fill a gap in your diary.
The prompt that does this arithmetic for you, and shows its working so you can check it:
I am a self-employed psychotherapist planning a closed 8-session group of 90 minutes each.
My figures: fee per member per session $45; room hire $25 per session; 10 screening calls of
20 minutes; 30 minutes of notes after each session; 3 hours of preparation; 2 hours of
supervision at $90 an hour.
Build a table of income, costs and effective hourly rate for 4, 5, 6, 7 and 8 members.
Show every calculation line by line. Then tell me the smallest group that beats my individual
rate of $110 per 50 minutes, and say clearly which numbers are my assumptions.
Check the output by recalculating one row by hand. Assistants usually do the multiplication right but occasionally forget a cost line you gave them, so count the lines against your list. The result is a minimum viable group size. That figure matters later, because it decides how many people you need to recruit and how long your waitlist must be.
The dropout cushion
People withdraw between agreeing and the first session. If you need six to run, plan to confirm eight. Ask the assistant to model a 20 per cent dropout between sign-up and session one, and a second figure for the first three weeks, then replace those guesses with your own history once you have it. The figures are assumptions, and you should label them as such in your notes.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowGet a programme outline you can edit, not accept
The clinical content comes from your training and your chosen model. What AI is good at is laying out structure: a session-by-session skeleton with timings, a standard opening and closing round, a handout list and a consistent rhythm. Give it your decisions and ask it to format them.
Act as an administrative assistant for a psychotherapist. I have decided the following and do
not want you to change any of it: closed group, 8 weekly sessions of 90 minutes, 6 to 8 adult
members, a psychoeducation-plus-process format, 10-minute check-in, 20-minute topic input,
40 minutes of group discussion, 10-minute close. The weekly topics are: (1) agreeing how we
work together, (2) noticing patterns, (3) boundaries, (4) anger and conflict, (5) shame,
(6) asking for what you need, (7) endings and change, (8) review and next steps.
Produce a one-page programme outline with timings for each session, a facilitator checklist
for the first ten minutes, and a list of handouts I would need. Do not make claims about
outcomes, and do not suggest clinical techniques. Mark anything you are unsure of.
An illustrative output, shortened: "Session 3: Boundaries. 0-10 check-in round. 10-30 short input on what boundaries are and why they differ between people. 30-70 group discussion with two prompts. 70-80 pair reflection. 80-90 close and one-sentence take-away. Handout: boundary reflection sheet." That is useful scaffolding. What you would fix: replace any generic discussion prompts with your own, strip any phrase such as "will learn to" or "will overcome", and check that the closing minutes leave time for a proper ending in a group where people may have become emotional.
Keep the outline as a document with your name on it, dated by version. If a member's complaint ever arises, a dated outline written by you and edited by you is a clear record of what you planned.
Screening questions: AI drafts the list, you read every answer
A screening form does two jobs: it gives you what you need to judge fit, and it tells the person what they are signing up for. AI can draft a clear form quickly. It must not score people or recommend who joins.
A sensible form for the group above might have five sections: contact details kept on your normal system, current support and any other therapist, what the person hopes to get from the group in their own words, practical availability for all eight dates, and how they feel about speaking in front of others. Ask the assistant to keep the wording plain and non-leading.
Draft a screening questionnaire of no more than 12 questions for people enquiring about an
8-week closed psychotherapy group. Plain English, reading age about 12, warm but not
clinical. Include: availability for all 8 dates, whether they are currently seeing another
therapist, what they hope for from a group in their own words, how they feel about sharing
in front of others, and whether they understand confidentiality limits in a group. Do not
ask for diagnoses or medication details. Do not include any question that implies the
group will treat a condition. End with a sentence explaining that I will contact them to
talk it through and that answering does not guarantee a place.
Two things to change in what comes back. First, check every question against what you actually need: if you would not use an answer, delete the question, because every extra field is more personal data you hold. Second, add your own risk questions in your own words, following your professional body's guidance and your practice's safeguarding procedure. AI should not write these, because getting them wrong has consequences that a draft cannot catch.
Summarising answers without exposing the person
Once forms arrive, you may want a one-page summary per person. If you do that with a general chat tool, you are sending personal and possibly sensitive information to a third party. Either do it by hand, or use a tool that your professional body, insurer and data-protection adviser have cleared, on a business plan whose terms say your content is not used for training. As a rule of thumb: drafting needs no personal data, summarising does. See the anonymising method if you are unsure what counts as an identifier. The longer discussion of what therapists can paste into chat tools is in this tutorial on ChatGPT and session notes.
Whatever tool you use, the screening call is the real decision point. The form tells you who to ring. The call tells you whether they fit.
The invitation email: describe, don't promise
Most group invitations fail because they describe the topic in abstract language and hide the practical facts. People decide on dates, fee, place and who else will be there. Put those first.
Here is a before and after, both illustrative. Before: "Join our life-changing healing group and finally break free from anxious patterns. Places limited!" After:
Subject: A weekly therapy group starting on [date]
Hello [first name], I am running a closed psychotherapy group for adults, meeting on [weekday] evenings from [time] to [time] for eight weeks starting [date]. It is a small group of six to eight people with me as the facilitator. The fee is $45 per session, payable in advance for the block. Before anyone joins, I speak to them for about 20 minutes to check that the group is a good fit and to answer questions. If you would like to find out more, reply to this email or use the short form at [link]. Answering the form does not guarantee a place. [your name]
Notice what the second version leaves out: no promise of improvement, no urgency, no claim about a condition. Ask the assistant to produce three subject-line variants and a version for referrers, then choose by reading them aloud. Check the final text against your professional body's advertising rules, which differ by body and are yours to read.
A short note for referrers
GPs, other therapists and community organisations refer more readily when they can see exactly who the group is for and what happens next. A half-page referrer note covers who the group suits, who it does not suit, dates and fee, how to refer, how you will confirm with the referrer, and what you will and will not share back. Ask for it in plain text with no promises. Send it individually, and write the personal opening line yourself.
Waitlist and confirmation sequence
Once enquiries start, the admin multiplies. A simple sequence keeps it tidy. Each stage is a template you reuse, and AI writes the template once; you send it with the personal details filled in by hand.
- Acknowledgement within a day: thanks, what happens next, the screening form link.
- Screening call booking: two or three time options.
- Offer of a place: dates, fee, payment method, cancellation and withdrawal terms, a deadline to confirm.
- Waitlist message: honest about numbers and timing, no pressure.
- Reminder one week before: dates, location or link, what to bring.
- Not-a-fit message: kind, short, and with other options you would stand behind.
The not-a-fit message is the one most therapists find hardest, so it is a good job for AI. Ask for a version that is brief, does not give reasons that could read as a clinical judgement in writing, and points to alternatives you name. Edit it until it sounds like you. If the person might be at risk, a template is the wrong tool: pick up the phone.
Keeping a waitlist that stays warm
Waitlists lose people quietly. Keep a simple sheet with first name or initials and a reference number, preferred days, date of last contact and status. Do not put clinical notes in it. Set a reminder to message everyone every three to four weeks with one honest line about timing. A second group starting is a better reason to contact the list than a generic newsletter. If your list is bigger than your first group, say so: telling people that a second block is likely, subject to confirmation, is kinder than silence.
A group agreement drafted from your decisions
Members usually sign something before the first session: a group agreement covering attendance, confidentiality between members, contact outside sessions, what happens if someone leaves, and payment. AI can turn your decisions into plain English faster than you can type them, which is the right use. It cannot tell you what your professional body requires, so treat the draft as a starting point for your supervisor or adviser to review.
A filled-in extract, illustrative only, from a prompt that supplied the rules and asked for plain wording:
Attendance: The group is closed, so everyone starts and finishes together. If you cannot attend a session, please tell me by [time] on the day. Missing more than two sessions means we will talk about whether the group still suits you.
Contact between sessions: Please do not arrange to meet other members outside the group during the eight weeks. If something comes up that you would like to raise, bring it to the group or to me.
Fees: The fee covers your place for all eight sessions, whether or not you attend each one. I do not refund missed sessions.
Read each line and ask whether you would actually enforce it. An agreement full of rules you will not apply is worse than a short one. Cut anything that is a generic default rather than your own decision, and make sure the fee and cancellation lines match what the invitation promised. If they differ, members will notice.
One mistake, and how it showed up
An illustrative case. A therapist planning a second group asked an assistant to "summarise these ten enquiry emails and rank who is most suitable". The emails contained names, workplaces and descriptions of personal difficulties. Two problems followed. The identifiable material had gone to a consumer tool the therapist had not checked against their insurer's conditions, and the ranking was presented as if it were a clinical view, which then anchored the therapist's thinking before any call took place.
The fix was to split the job. The therapist now drafts generic templates with the assistant, handles every enquiry personally, and keeps a plain sheet of dates, outcomes and reference numbers. Summaries, if any, are written by the therapist after each call. That costs about ten minutes per enquiry, which is cheaper than an uncomfortable conversation with a regulator or a member.
Walking a launch through, week by week
Here is a compressed timeline for the six-to-eight-member group above, with AI's part marked. These are suggested timings, not rules.
| When | What you do | AI's part |
|---|---|---|
| Week -6 | Decide model, dates, fee, minimum size | Fee and capacity table |
| Week -5 | Write outline; speak to supervisor | Format and timings |
| Week -4 | Publish invitation; email referrers | Drafts of both |
| Weeks -4 to -2 | Screening calls; offers; waitlist | Templates only |
| Week -1 | Confirm, reminders, consent forms | Reminder wording |
| Week 1 onward | Run the group; notes by your usual method | None for client content |
The pattern is that AI touches documents with no personal data in them, and you handle every conversation. If the group falls short at week -2, the decision is yours: run with five if the numbers still hold, postpone, or merge with a second block. Use your table from the start to see which choice you can afford.
Errors that show up in drafts, and how to catch them
- Outcome language creeping in. Assistants like phrases such as "build resilience" or "heal". Search the draft for verbs of change and rewrite as description.
- Invented specifics. A draft may mention a "manualised programme" or a named model you never specified. Delete anything you did not supply.
- Fee arithmetic drift. Recheck any total against your own calculator, particularly when you change group size.
- Over-friendly tone to anxious people. Some drafts push urgency ("Don't miss out"). Remove it.
- Confidentiality wording that overstates. Group confidentiality depends on every member, not only you. Ask your supervisor or professional body for wording rather than accepting a draft's promise.
A quick check that the workflow worked: after a month, count enquiries, screening calls, offers, confirmations and first-session attendees. The step where most people disappear tells you what to change. If it is between enquiry and call, your reply is too slow or the form too long. If it is between offer and confirmation, the fee or dates are the issue.
For what happens after the group begins, such as notes and consent across members, read AI for group therapy practices. If you run this inside a bigger service, the group-practice roll-out plan is the better companion. The hub for everything else is the AI guide.
The never-paste list for group enquiries
Keep a short list taped to the desk. Names, contact details, dates of birth, employers, referrer names, anything a member said in a call, any clinical impression, and any screening answer in identifiable form. If a draft needs an example person, use an invented one and say it is invented. If you decide to use a tool that does process identifiable information, make that decision with your insurer, professional body and a data-protection adviser, and write down what you decided and why. That record is more useful than any tool.
Finally, consent. Anything you tell members about how their information is handled should be in your privacy notice and in the group agreement they sign. If AI features anywhere in handling their information, say so plainly. If it does not, you can say that too. Either answer is fine; a vague one is not.
Questions therapists ask before using AI for a group
Can AI decide who is suitable for my group?
No. AI can tidy screening answers into a one-page summary, but suitability is a clinical judgement that stays with you. Treat any suggestion it makes as a prompt for your own questions, and record your decision and reasons yourself.
Should I paste a prospective member's enquiry into ChatGPT?
Not in a form that identifies them. Strip names, contact details, employer and anything distinctive, or keep identifiable material inside a tool your insurer, professional body and data-protection adviser have approved. Drafting generic invites and outlines needs no personal data at all.
How many members should a new group start with?
That depends on the model, the population and your training, so check your professional body's guidance. Many clinicians plan to recruit a few above the minimum they need to run, because people drop out before session one. The sums below show how to test your own numbers.
Can the invite email say the group will help with a condition?
Avoid outcome promises. Describe what the group offers, who it is designed for, the structure and the fee. Say what it is, not what it will fix. Your professional body's advertising rules govern the exact wording, so read them before sending.
Do I need to tell members I used AI?
If AI touched anything about them, such as a summary of their screening answers, say so in your privacy notice and consent form. If you only used it to draft a generic outline and email template, there is nothing about them to disclose, though many clinicians mention it anyway.
Further reads
- Should Therapists Tell Clients They Use AI? — Wording for telling clients how AI is used in your practice.
- AI Note-Taking Tools for Therapists: A Confidentiality Checklist — A checklist for judging any tool that sees client information.
- Best AI Software for Therapists in a Group Practice: Seat Costs — Seat costs if the group sits inside a larger practice.
- Telling Patients You Use AI Note-Taking: Consent Wording That Works — Consent wording that works when AI touches notes.
- AI in Psychotherapy Practice: Enquiry Replies and Fee Policies — An admin-only template workflow for first enquiries, fees and cancellations, with a crisis hand-off rule.
- Private Practice Policy and Procedure Manual: A Therapist's AI Draft — Draft a therapy practice policy manual section by section with AI, then check it with a professional and set a review date.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI business data page and Anthropic privacy centre article on model training (both checked 4 November 2026: business and commercial plans are not used for training by default); professional-body advertising and ethics guidance to be checked by the reader.