ChatGPT Fitness Coach Prompt: Client Programme Template for Trainers

Use a fixed prompt with four parts: the client's intake answers, hard safety rules (no diagnosis, no rehab advice), a red-flag gate that returns "refer" instead of a programme, and a set output format. ChatGPT drafts a four-week block; you check every exercise, set count and time before the client sees it.

That review is the point. ChatGPT has never watched this client squat, so the draft is a first pass from a text tool. OpenAI's usage policies draw the same line for anyone building or offering AI tools: no tailored advice that requires a licence, such as medical advice, without a licensed professional involved. Diagnosing or treating an injury is that kind of advice, so it stays out of the prompt. The prompt below is built to make your review quick: it states its assumptions, lists what it could not check, and stops when the screening answers say it should.

Follow me on Instagram@sagnikteaches

The scope here is the prompt and the review step, for adult clients you have already screened. Building the intake form is covered in the tutorial on trainer intake forms and health data, and the wider question of where ChatGPT belongs in coaching sits in how personal trainers can use ChatGPT safely for client programmes.

Connect on LinkedInSagnik Bhattacharya

What goes into the prompt, and what stays out

Start with the fields, because the draft is only as good as the intake behind it. A programme needs eleven things: a client label, an age band, one measurable goal, training history, days and session length, equipment, the movements you have watched the client perform, the movements to avoid or modify, the screening result, resolved injury history, and likes and dislikes. If your intake form does not capture them, fix the form before you touch the prompt.

Subscribe on YouTube@sagnikteaches

Leave out the surname, phone number, address, date of birth, medical records, and any letter from a doctor or physiotherapist. A label such as "Client A", an age band, and a restriction written in your own words ("no loaded spinal flexion this block") give ChatGPT everything it needs to work around a limitation without holding the sensitive record. A physiotherapist's note stays in your client file, and you translate it into one line.

Three settings are worth checking before the first client. On a personal plan, switch off "Improve the model for everyone" under Data controls; business plans such as ChatGPT Business do not train on business content by default. For one-off drafts you can use Temporary Chat, which stays out of your history, does not create or update memories, and is not used to improve models while it stays temporary. OpenAI says it may keep a copy for up to 30 days for safety, so it is not a vault. And decide now where the final programme lives: in your own client record, never only in the chat. The checklist for client health data and AI tools covers the data-protection side; ask a qualified adviser about the rules that apply to you.

The prompt: a red-flag gate first, then intake, rules and output

Paste this into a new chat or into a Project's instructions, then fill the square brackets for each client. The gate list is an example, not a medical standard: replace it with the rules from your own screening form and your certifying body's guidance.

ROLE
You are helping a qualified personal trainer draft a training programme for one adult client, identified only by the label below. You are not the client's coach, doctor or physiotherapist. The trainer will review and change everything before the client sees it.

STEP 1: RED-FLAG GATE (do this before anything else)
Read the SCREENING and INJURY HISTORY lines. If any of these is present, write only "REFER BEFORE PROGRAMMING", list each flagged item, and stop. Do not draft a programme.
- any "yes" on the screening form
- current pain, or pain during any movement the trainer has watched
- chest pain, dizziness or fainting during or after exertion
- pregnancy, or a birth in the last 12 months
- surgery or an injury in the last 6 months
- medication that affects heart rate or blood pressure
- a diagnosed heart, lung, kidney, metabolic or neurological condition
- anything marked "unsure"

STEP 2: INTAKE
Client label: [Client A]
Age band: [40-49]
Goal, one measurable sentence: [...]
Training history: [...]
Days and session length: [...]
Equipment available: [...]
Movements the trainer has watched and rated safe: [...]
Movements to avoid or modify (trainer's words): [...]
SCREENING: [...]
INJURY HISTORY (resolved, no current symptoms): [...]
Likes and dislikes: [...]

STEP 3: RULES
- Do not diagnose, name a condition, or suggest exercises to treat, rehabilitate or "fix" pain or an injury.
- If the programme would conflict with a "movements to avoid" line, say so instead of working around it.
- Use only the equipment listed. Prefer common, low-skill exercises.
- Set effort with RPE (6 = about four reps left in reserve, 7 = about three). Never use percentages of a one-rep max.
- Change only one variable (sets, reps or effort) from one week to the next. Week 4 is a lighter week.
- Assume 40 seconds of work per set plus the rest you list, and show the session time as a sum.
- No promises about results, weight loss or body shape.
- If information is missing, do not guess. Ask up to five questions at the end.

STEP 4: OUTPUT, in this order
1. Gate result: PASS or REFER, with reasons.
2. For each training day, a table: exercise, sets x reps, effort (RPE), rest, one coaching cue, an easier swap using the same equipment.
3. A week-by-week progression table for weeks 1 to 4.
4. Hard sets per week for each major muscle group, counted line by line.
5. Estimated session time, with the sum shown.
6. Assumptions I made. Questions for the trainer. Things I could not check.
Write in plain language the client could read.

Each block earns its place. The gate sits first so that the stop instruction is the first thing the model reads, and you should test it (the last section does) rather than assume it works. The PAR-Q+ site describes seven general health questions on its first page, with follow-up questions for anyone answering yes to one or more, and it warns against outdated or altered copies, so use the official form if you do not already have a screen.

In the rules block, two lines do most of the safety work. "Do not diagnose" keeps the draft inside exercise selection. "If the programme would conflict with a movements-to-avoid line, say so" matters because a helpful-sounding model will quietly engineer around a restriction to give you something usable. You want the conflict printed where you can see it. RPE replaces percentages because a percentage of a one-rep max needs a tested max, and a client back from 18 months off will not have one. The single-variable rule stops the common draft habit of adding sets, reps and effort in the same week.

The output list forces two habits. Counting hard sets line by line and showing the session time as a sum give you numbers to recount, which is how you catch arithmetic slips. And the last item, assumptions, questions and things it could not check, is where your review starts, because it tells you which parts the model itself is unsure about.

Client A: a returning lifter with a past back flare-up

Say a trainer takes about 40 minutes to write a first block from a blank page, and wants to cut that without cutting corners. Client A is a made-up adult who lifted for two years in their late twenties, has done nothing structured for 18 months, and sits at a desk all day. Here is the filled-in intake, as the trainer would paste it under the gate and rules.

Client label: Client A
Age band: 40-49
Goal, one measurable sentence: Do 10 push-ups from the toes and jog for 20 minutes without stopping by the end of week 12.
Training history: Lifted three days a week for about two years in their late 20s; no structured training for 18 months; desk-based job.
Days and session length: 3 days a week, 45 minutes including warm-up, non-consecutive days.
Equipment available: Commercial gym with barbells, dumbbells to 40 kg, cable stack, leg press, lat pulldown, benches.
Movements the trainer has watched and rated safe: Goblet squat, dumbbell Romanian deadlift, incline push-up, seated row (assessment session, no pain reported).
Movements to avoid or modify (trainer's words): No loaded spinal flexion or extension this block. Keep hinges to mid-shin depth.
SCREENING: All seven general health questions answered no. Trainer rates the client ready for low-to-moderate intensity.
INJURY HISTORY (resolved, no current symptoms): One episode of lower-back pain about two years ago; resolved, no treatment now.
Likes and dislikes: Likes cables; dislikes burpees and treadmills.

Notice what the intake does not contain: no name, no date of birth, no doctor's letter. It has one measurable goal rather than "get fitter", a note of which movements the trainer has actually watched, and a restriction in the trainer's own words. Nothing in it trips the gate, so the draft proceeds.

What ChatGPT returned, and the tally to recount

An illustrative draft of Day A comes back as a table, with a gate line above it reading "PASS: no flagged items".

Days B and C follow the same shape: leg press, seated dumbbell shoulder press, lat pulldown, hip thrust and dead bug on B; split squat, incline push-up, one-arm dumbbell row, cable pull-through and bird dog on C. Then the progression table, and a note that Day C in week 3 swaps the goblet squat for a barbell back squat at 75% of the client's one-rep max.

Below that, the draft's own weekly tally reads: quads 12 hard sets, back 12, chest 9. And its session estimate says "about 45 minutes". The tally and the time estimate are both wrong, and one of the exercises breaks a rule the trainer wrote into the intake. The assumptions list is honest about some gaps (an 8-minute warm-up on a rower, every station free at once, a client who can hold a 20-second plank) and its questions ask whether the client prefers a machine or free-weight squat, and whether the cable stack is usually busy at the session time. It flags no conflict with "no loaded spinal extension", which is the failure to look for.

The review: nine checks, run on Client A's draft

1. Gate against the paper. Match the screening line to the signed form, not to what you typed. "All seven general health questions answered no" has to be true on the form itself. If you typed it from memory, stop and look.
2. Equipment. Tick every exercise against the equipment list. The intake names benches, a cable stack and a leg press but no back extension station, so ChatGPT assumed one exists.
3. Recount the sets. From the tables: quads are 3 + 3 + 3 = 9, back is 3 + 3 + 3 = 9, chest is 3 + 3 = 6. The draft's own tally said 12, 12 and 9. This trainer's rule for a returning client's first block is 6 to 12 hard sets per major muscle group a week, and all three groups sit inside it, but any decision made from the draft's tally would have been wrong.
4. Sum the time. Day A as drafted has 18 sets at about 130 seconds each (40 seconds of work plus 90 seconds of rest), which is 2,340 seconds or 39 minutes. Add the 8-minute warm-up and the session is 47 minutes before the client walks between stations, against the 45 allowed.
5. Progression. Week 3 raises sets (3 to 4), reps (8 to 10) and effort (RPE 7 to 8) together, which breaks the one-variable rule in the prompt. The week-3 back squat at 75% of a one-rep max needs a max nobody has tested.
6. Restriction conflicts. Read each exercise against the "avoid or modify" line. A back extension conflicts with "no loaded spinal flexion or extension this block", and the draft did not say so.
7. Goal coverage. Trace each part of the goal to something in the plan. Client A's goal has two parts, 10 push-ups from the toes and a 20-minute jog, and the draft trains neither directly: Day C has incline push-ups and there is no conditioning at all. Either add a push-up progression and some walk-jog intervals outside the gym (the client dislikes treadmills), or agree with the client that this first block builds strength only and write that into the programme. If you see this gap often, add "Show which exercises train each part of the goal" to the output list.
8. Claims. Strike any sentence that promises a result or explains the client's body to them. The draft ends with "Stick with this and you'll build a stronger, healthier back", which is a promise to a client with a back history.
9. Sign-off. Save the final programme and the intake it came from in your own client record, marked "Reviewed and approved by [your name], [date]". The chat is not the record.

In this made-up case the review takes about 15 minutes, plus 5 to fill the intake and read the draft, so roughly 20 against the 40 from a blank page. Treat those as placeholders and time your first five clients. The other number to track is how often you make the same edit. After ten clients, any correction you keep making belongs in the rules block, so the next draft arrives already fixed.

When the gate fires: Client B should not get a programme

A second made-up client. Client B is in their fifties, wants to lose weight before an event six months away, has not exercised for five years, takes medication for blood pressure, mentions feeling dizzy when standing up quickly, and answered yes to two of the general health questions. The intake looks complete, and a keen model would happily write a four-day plan. With the gate first, the reply should be short.

REFER BEFORE PROGRAMMING
Flagged items:
1. "Yes" answers on the general health questions (two).
2. Medication that affects blood pressure.
3. Dizziness on standing.
I have not drafted a programme. Next step for the trainer: follow your screening process before any programming.

That is the whole answer, and it is the right one. What happens next belongs to you and the client: follow-up screening questions and, where your process calls for it, written clearance from the client's doctor, before anyone plans a session. The model has no business suggesting what is wrong or what is safe.

Put the gate at the bottom of the prompt, after the output format, and you invite a different result: a complete four-day programme with a line at the foot saying "consult your doctor before starting". That looks responsible and is not, because the programme now exists and can be handed over. A softer instruction such as "be careful with clients who have health conditions" has the same flaw, since it leaves the model to decide what counts. The gate works because it names the triggers and says stop.

Two follow-up tests are worth running. Delete the dizziness line and send Client B again: the blood-pressure medication and the yes answers should still trigger a referral. If they do not, tighten the wording. And remember the limit of any gate: it only sees what you wrote down. If a client mentions knee surgery in conversation and you do not add it to the intake, nothing fires. Update the intake itself, not just your memory.

Four edits for clients who are not Client A

• A time-poor client with two 30-minute sessions. Change the days line to "2 days a week, 30 minutes including warm-up" and add a rule: "Maximum four exercises a session, paired as supersets (two exercises back to back, then one rest)." The sum shows why. Two exercises done separately, 3 sets each at 130 seconds a set, take 13 minutes. Paired, 3 rounds of 40 + 40 + 90 seconds take 510 seconds, or 8.5 minutes. Two pairs are 17 minutes, so with a 5-minute warm-up and a 3-minute cool-down the slot has about 5 minutes spare.
• Home training with one pair of adjustable dumbbells and a resistance band. Replace the equipment line and add: "Describe load by RPE, because the dumbbell steps are coarse. Every exercise needs a swap that uses the same equipment. No bench, rack or cable exercises." On review, check each exercise against that list; drafts drift towards gym equipment, so a bench press can reappear.
• A goal that is not measurable, such as "tone up". Rewrite it before pasting: ask what the client would be able to do in 12 weeks that they cannot do now, and turn it into something you can test, such as eight incline push-ups and a 45-second plank by week 8. A vague goal makes ChatGPT guess what the client wants, and it guesses in its own words.
• A client with written restrictions from a physiotherapist. The letter stays in your file. In "movements to avoid or modify" write your own one-line version, for example "No loaded spinal flexion until the next review in six weeks". If the restriction follows a recent injury or surgery, the gate's six-month line fires by design. Once you hold the physiotherapist's written clearance and your own process accepts it, you can edit that gate line for this client on purpose and note why in the record. The model never makes that call.

Where to keep it: a Project, a chat per client, and your own record

Create a ChatGPT Project called "Client programmes" and paste the role, gate and rules blocks into its instructions, leaving the intake and output blocks to paste per client. Project instructions apply only inside that project and override your global custom instructions, so your other chats do not inherit them. Use one chat per client, named with the label rather than the name. The tutorial on using ChatGPT Projects to keep client work separate covers the set-up in more detail.

Check memory before the first client. OpenAI describes a project-only memory setting under which a project's chats do not reference your wider saved memories or conversations outside the project, and shared projects always use it. You will not need file uploads for this routine, because the template is text, and as of October 2026 file limits are 5 on Free, 25 on Go and Plus, and 40 on Pro, Business and Enterprise if you later add your exercise library. Keep client documents out of the project either way.

For occasional use without a Project, paste the whole template into a Temporary Chat; the guide to when staff should use Temporary Chat explains its limits. In both cases the final programme goes into your own system: a PDF or your coaching app, with the intake it came from and the sign-off line. Chats can be deleted, and temporary ones disappear. Also ask your insurer in writing whether drafting programmes with AI changes anything about your cover, and keep the reply with your records.

Test the prompt on four dummy clients before using it live

Give the template a week of dummy intakes before a real client sees any output. Each test has a pass mark.

1. A clean client. Everything screens clear. Pass: a gate line saying PASS, then all six output items in order, with the sets counted and the session time shown as a sum. Fail if any item is missing or the sum is absent.
2. A single red flag. Make everything clear except one line, such as blood-pressure medication. Run it three times. Pass: all three runs return REFER and nothing else. If one run returns a programme, tighten the gate with "even if every other answer is clear".
3. A conflict. Put "no loaded spinal extension" in the avoid line, and "likes back extensions" in the likes line. Pass: the draft prints the conflict and leaves the exercise out. Fail if it quietly includes it or quietly omits it without saying.
4. Missing information. Delete the equipment line. Pass: up to five questions at the end and no invented equipment.

Keep a one-line log: the date, the model name shown in the picker, and pass or fail for each test. Model names and behaviour change often, so re-run the four tests whenever the picker changes or you edit the template. A prompt that passed in October may behave differently by the new year, and the log tells you when it started.

What this routine buys you is a faster, more consistent first draft and a built-in prompt to look harder at the parts most likely to be wrong. What it does not buy is a substitute for your judgement of the person in front of you. The numbers in this tutorial are made up, so treat every figure as a prompt to run your own: count your own sets, time your own sessions, and tighten the rules wherever you find yourself making the same edit twice.

Questions trainers ask before using the prompt

Can I use a ready-made fitness coach GPT instead?

I would not. OpenAI is retiring custom GPTs, and existing ones, including other people's coaching GPTs, stop running on 11 December 2026. You also cannot control or audit someone else's instructions. Paste the template into a ChatGPT Project's instructions instead, so the rules are yours and stay visible.

Should I paste the client's name or medical history?

No. Use a label such as Client A, an age band and restrictions written in your own words. Keep the intake form, any doctor or physiotherapist letter and the signed programme in your own client record, and check what data-protection rules apply to health information where you work.

Can I give clients the prompt to use at home?

Not as written. The gate depends on your screening and your review, and a client pasting their own health details into a consumer chat gets neither. If clients ask about using AI between sessions, point them to a short check-in routine you control instead.

How often should I re-run it for the same client?

At the start of each new block, with a fresh intake, because screening answers, injuries and equipment change. Within a block, adjust by hand. Re-running mid-block gives a different draft that may not match what the client has already learned.

Further reads

• Personal Trainer Progress Reports: Monthly Client Summaries With AI — Turn each finished block into a short monthly client report.
• AI Prompts for Personal Trainers: Session Recaps and Renewal Texts — Prompts for session recaps and renewal texts once programmes are running.
• Best Free AI for Personal Training: Client Programmes and When to Pay — What the free tiers allow before you need a paid plan.
• AI Agent for Personal Trainers: Check-Ins and Where It Must Stop — Where automated check-ins must stop and hand back to you.
• ChatGPT Apps for Fitness Studios: Seven Jobs and Their Limits — Seven studio jobs for ChatGPT's connected apps, and where they stop.
• Can AI Draft a Personal Trainer Contract? What a Solicitor Must Check — Which parts of a trainer contract need a solicitor's eye.
• AI for Fitness Trainers: Client Check-Ins Without Medical Advice — Weekly check-ins that never drift into medical advice.
• AI Software for Personal Trainers: Scheduling, Check-Ins, Billing — Free against paid tiers of Trainerize, Everfit and TrueCoach, which AI features earn their fee, and why you stay accountable for every plan.
• AI Fitness Video Maker: Turn a Class Plan Into Member Videos — A trainer's workflow from written class plan to short member videos, with the safety and music-licence checks that matter.
• AI Nutrition Advice at a Gym: What Is Safe to Offer Members? — A gym-owner policy for AI nutrition content: general information versus personal advice, safe claims, disclaimers and referral triggers to a dietitian.
• Can AI Write CrossFit Programming? Scaling Options and Coach Notes — A coach's guide to AI-drafted scaling ladders and whiteboard notes: the prompt, a worked draft with its errors marked, and seven checks before anything goes up.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: PAR-Q+ official site (general health questions and follow-up route); OpenAI usage policies (updated October 2026); OpenAI help pages on Projects and Temporary Chat; OpenAI's custom GPT retirement notice; ChatGPT plan details from OpenAI's help pages (October 2026).