Use ChatGPT to draft and vary programmes from your own assessment, never as the assessor. Leave out names and identifying details, describe injuries and limits in general terms, give it the equipment and experience level, and check every exercise, load and progression before the client sees it. Anything medical goes to a clinician, not the chatbot.
The risk isn't that ChatGPT writes a bad programme; it's that it writes a plausible one. A confident, well-formatted plan with a contraindicated exercise or a progression too steep for a deconditioned client looks exactly like a good one. Everything below keeps you as the professional in the loop: what to type and what to leave out, a prompt that builds your constraints in, a five-minute check, a reusable setup, and wording for telling clients.
Drafting versus prescribing: where the line sits
| ChatGPT is useful for | Keep it away from |
|---|---|
| Structuring a block: a four-week strength phase, three days a week, 50-minute sessions | Deciding whether someone is fit to train. That's your screening, and a doctor's clearance where needed |
| Swapping exercises for equipment, preference or a stated limit | Diagnosing pain or designing rehab for an injury |
| Progression options and deload weeks | Programming around medical conditions, pregnancy or recent surgery beyond your qualification and without clinician input |
| Client-friendly explanations of why an exercise is in the plan | Diet plans for medical conditions or disordered eating |
| Warm-ups, finishers, session timings, home versions | Answering a client's "is this pain normal?" |
The right-hand column is about your scope of practice, not about AI. The same boundaries apply whether you write the plan by hand or not; AI just makes it easier to drift over them because it will cheerfully answer any question. Your certifying body's guidance on scope, and your insurance policy's wording, are the documents to check if you're unsure where your line is.
What to type, what to generalise, and what never to enter
| Type freely | Generalise first | Never enter |
|---|---|---|
| Goal, training age, days available, session length, equipment, exercises they enjoy or hate | "Left knee: physio-cleared for loaded knee flexion to about 90 degrees; avoid deep lunges" instead of a diagnosis and history | Name, date of birth, contact details, photos, medication names, the screening form itself |
| Your programming rules and preferred format | "Blood pressure managed; GP has cleared moderate training" instead of readings and prescriptions | Anything a client told you in confidence about their health or life |
Health details count as special category data under data-protection law such as the GDPR, which is why the middle column exists: you translate the clinical picture into a training constraint yourself, and only the constraint goes into the chat. Use a client code ("Client 14") rather than a name.
A before and after makes the translation concrete. Before, the notes a busy trainer might paste straight in: "Woman, 52, works at [employer], ACL reconstruction on the left knee a few years ago, still some swelling after long walks, on medication for blood pressure, last reading a bit high." After, the version that goes into ChatGPT: "Client 14, 50s. Left knee: cleared for loaded knee flexion to about 90 degrees; keep impact low and review after higher-volume leg days. Blood pressure managed; cleared for moderate training; avoid long breath-holds under heavy load." The second version gives the model everything it needs to programme around, and nothing that identifies the client or describes her medical history. The judgement about what the swelling and the reading mean for training stayed with you, which is where it belongs.
Then the settings. On a personal ChatGPT plan, switch off the model-training option in privacy settings. For one-off drafts, OpenAI's Temporary Chat keeps the conversation out of your history and out of model training, though OpenAI may keep a copy for up to 30 days for safety purposes. On ChatGPT Business, your content is left out of model training by default. The broader rules for any team handling client data are in keeping customer data private when your team uses AI.
A programme prompt that builds your constraints in
Most bad AI programmes come from thin prompts. Put your professional decisions into the prompt so the model is filling in a structure you've already set:
You're helping a qualified personal trainer draft a programme.
The trainer has assessed the client; you do not make safety decisions.
Client 14: goal = build strength for hiking; training age = 1 year;
3 sessions a week, 50 minutes each; commercial gym, full equipment.
Constraints (from my assessment): avoid deep knee flexion past ~90
degrees on the left knee; no overhead barbell pressing (shoulder
comfort); enjoys kettlebells, dislikes running.
Write a 4-week block. For each session give a table:
exercise | sets | reps | effort (RPE) | rest | coaching note.
Progression rule: add reps within the range before adding load;
no more than one variable changes per week.
Week 4 is a deload at about 60% of week 3 volume.
Only use common, clearly named exercises.
List any exercise you're unsure is suitable given the constraints,
and ask me questions if anything important is missing.The last two lines do a lot of work. Asking the model to flag its own uncertainty and to ask questions surfaces the gaps you'd otherwise find only when checking.
An illustrative extract of what comes back for Session A in week 1:
| Exercise | Sets x reps | RPE | Rest | Coaching note |
|---|---|---|---|---|
| Goblet box squat | 3 x 8-10 | 6-7 | 90 s | Box set so the knee stays at about 90 degrees |
| Kettlebell Romanian deadlift | 3 x 10 | 7 | 90 s | Hinge from the hips, soft knees |
| Bulgarian split squat | 3 x 8 each side | 7 | 60 s | Great for hiking-specific single-leg strength |
| Half-kneeling kettlebell press | 3 x 8 each side | 7 | 60 s | Press overhead in a straight line |
| Weighted step-up | 3 x 10 each side | 7 | 60 s | Builds strength for climbs |
It closed with one flag ("confirm the step-up box height suits the knee") and one question ("are there any hikes with heavy packs coming up?"). Useful, but the draft still needs three changes. The Bulgarian split squat takes the front knee well past 90 degrees on most people, so it breaks the left-knee constraint without the model noticing. The kettlebell press is overhead pressing with a different implement: your note said "no overhead barbell pressing" for shoulder comfort, and the model read it literally. Only you know whether a light kettlebell press is fine or whether the note should have said "no overhead pressing". And the week-2 table (not shown) added a set and 2 kg at once, against the one-variable rule. None of these are unusual. They're the reason the next step exists.
The five-minute check before a client sees it
Every draft, every time. This is the step that makes AI use safe, so don't let it shrink when you're busy:
- Every exercise is real and correctly named. Models occasionally invent hybrids or mislabel variations.
- Nothing on the avoid list, including disguised versions. "Avoid deep knee flexion" should also rule out pistol squats and deep split squats, which a model may not connect.
- Weekly volume per muscle group sits within your usual range for this client's level. Drafts often overload beginners by listing too much. A quick count settles it: in the draft above, Session A has 12 sets for the legs once the split squat is swapped for a single-leg Romanian deadlift, and if Session C adds leg press 3 x 12 (to 90 degrees) and hamstring curls 3 x 12, that's 18 hard leg sets in the week. If your range for a client at this level is, say, 10-14, cut Session C back before sending rather than hoping they'll skip sets.
- Progression follows your rule and isn't steeper than you'd write by hand.
- Timing: warm-up, working sets and rest actually fit the session length.
- Equipment matches what the client really has access to.
- Coaching notes are accurate. AI biomechanics explanations can be fluent and wrong.
- Anything touching their condition is within what their clinician cleared.
Note the date and that you checked it in the client's file. A filled-in entry needs only a line or two: "Client 14, block 3 (weeks 9-12). AI draft checked 27 Sep. Removed Bulgarian split squat (left knee), swapped kettlebell press for landmine press (shoulder), cut weekly leg volume from 18 to 13 sets, rewrote two coaching notes. Sent 28 Sep." That record matters if a client is ever hurt and asks how the plan was built. Setting up human review for AI work covers building checks like this into a routine that doesn't slow you down.
Where ChatGPT programmes typically go wrong
- Forgotten constraints. A limit mentioned early in a long chat may be ignored twenty messages later. It typically shows up like this: the block is fine, then in week 3 you ask "make Session C more hiking-specific", and the reply adds weighted-vest walking lunges and deep step-downs, both outside the knee limit you gave at the start. Restate the constraints block in every request, and start a new chat for each block.
- Everything is 3 x 10. Without your rules, drafts default to generic set and rep schemes. Give it your system.
- Confident injury talk. "This exercise will strengthen your knee and prevent further injury" is a claim you shouldn't pass on. Strip therapeutic promises from client-facing notes.
- Nutrition numbers stated as fact. Calorie and protein targets appear unasked and look precise. Remove them unless you'd stand behind them yourself.
- Made-up sources. Ask for the research behind an approach and you may get references that don't exist. Why AI makes things up, and how to catch it explains the mechanism.
Lower-risk jobs worth handing over first
If you're new to this, start with the work around the programme rather than the programme itself. These carry little risk because you're the one who decided the content:
- Explaining the plan to the client. Paste your finished programme and ask for a one-page, plain-English explanation of what each week is for. Check it, then send it. An illustrative line from one such explanation: "Week 1 builds the base. The box squats and step-ups will strengthen your knee and protect it from future injury on the trail." Keep the first sentence; change the second to "The box squats and step-ups build leg strength for climbs, at a depth your knee is comfortable with." The original promises an outcome for an injury, which is a claim you can't back and shouldn't put in writing.
- Travel and home versions. "Rewrite session B for a hotel gym with dumbbells up to 20 kg and a bench, keeping the same intent and the same constraints."
- Session notes. Dictate rough notes after a session and have them tidied into a consistent format for the client file, without names in the prompt.
- Admin: onboarding emails, policy wording, social posts about your training approach.
Once you trust your checking routine on these, moving to full programme drafts is a smaller step.
When a client arrives with a ChatGPT programme of their own
Some clients now arrive with plans they've generated themselves. Dismissing them outright tends to go badly: the client feels judged and the plan often has sensible parts. Go through it together instead. Keep what fits their goal, explain the changes in terms of their constraints ("this has deep lunges every session; with your knee, we'll use step-ups to a box instead"), and point out what the chatbot couldn't know, such as how they moved in your assessment. Handled well, it's a clear demonstration of what a trainer adds that a generated plan can't.
A reusable setup: one Project, your rules, no client files
ChatGPT's Projects keep related chats, files and instructions together, and a project's instructions apply only inside that project. Create one called "Programming" and put in it:
- Your programming principles: progression rules, volume ranges by level, deload approach.
- Your exercise library, if you keep one, so drafts use names you recognise.
- Your output template, so every programme arrives in the same format for your coaching app.
A filled-in set of project instructions, as an illustration of one trainer's rules rather than a recommendation: "You draft programmes for a qualified trainer who makes every safety decision. Format: one table per session with exercise, sets, reps, RPE, rest, coaching note. Progression: add reps to the top of the range before adding load; change one variable per week. Weekly hard sets per muscle group: 8-12 for clients under a year of training, 12-16 above that. Deload every fourth week at about 60% of the previous week's volume. Use only exercises from the attached library. Never write injury, rehab or nutrition advice. Flag anything you're unsure fits a stated constraint." Every rule in there is one you'd otherwise retype, or forget to, in every chat.
Keep client files out of it. Each block gets a new chat inside the project with a coded brief. Save your best briefs and prompts somewhere the whole team can reuse them; building a shared prompt library shows a simple structure.
Tools built for trainers
If you already deliver programmes through a coaching app, check its own AI first. ABC Trainerize's AI Workout Builder uses the client's goals, training history, preferences and past performance, is in Beta for coaches on its Grow plan and above, and produces workouts you can edit before assigning. The advantage is that the data stays in the platform you already use for clients (check its privacy terms); the trade-off is less flexibility than a general assistant.
On cost, as of September 2026: ChatGPT Plus is $20 a month. ChatGPT Business is $25 per user a month billed monthly, or $20 billed annually, with a minimum of two seats. For a solo trainer, Plus with training switched off and generalised inputs is the realistic setup; Business makes sense for a studio with several trainers.
Time saved for a trainer with 18 clients
Illustrative figures. Say a trainer writes a new four-week block for each of 18 clients every month, at about 40 minutes each: 12 hours a month.
- With ChatGPT: 10 minutes writing the coded brief, a few minutes for the draft, 10 minutes checking and editing. About 25 minutes each, or 7.5 hours a month.
- Time saved: about 4.5 hours a month, for $20.
- What doesn't change: the assessment, the constraint translation and the check. If the check shrinks to one minute to save more time, the savings come with risk attached.
One good use of the saved time is more frequent contact with clients between sessions; automated check-ins for gym members has timings and templates that work for a PT client list too.
Telling clients you use AI
Be open about it before anyone asks. A line in your onboarding pack covers it:
"I sometimes use AI tools to help draft the structure of your programme. Every exercise, load and progression is chosen and checked by me, and I never share your name or health information with those tools."
If that sentence would be untrue for the way you currently work, change the way you work rather than the sentence.
ChatGPT questions personal trainers ask
Is ChatGPT Plus enough, or do I need a business plan?
For a solo trainer, Plus at $20 a month is usually enough, provided you switch off the model-training setting in privacy settings and keep names and health details out of your prompts. ChatGPT Business costs $25 per user a month billed monthly, needs at least two seats, and doesn't train on business data by default. It suits a studio with several trainers sharing prompts rather than one person working alone.
Can I paste a client's PAR-Q or medical form into ChatGPT?
Don't. Those forms hold health information that deserves the highest care, and ChatGPT doesn't need it. Read the form yourself, decide what it means for training, and pass on only the practical limit in general terms, such as 'avoid overhead pressing; cleared for light pulling'. If the form raises anything that needs medical clearance, that decision belongs to a clinician, not to you or the chatbot.
Can ChatGPT write meal plans for my clients?
It can draft general healthy-eating guidance, but what you're allowed to give depends on your qualification. Calorie prescriptions, plans for medical conditions, and anything for clients with a history of disordered eating belong with a registered dietitian or doctor. If you do share general guidance, check every number yourself, because AI tools state calorie and protein figures with more confidence than the evidence usually supports.
Does using AI change my insurance cover?
Ask your insurer directly, because policies differ. What most insurers care about is that you worked within your qualifications and exercised professional judgement. Keeping a short record of your checks on each programme, and never sending an AI draft unreviewed, puts you in a much better position if a client is ever injured and asks how their programme was put together.
Further reads
- Is It Safe to Put Customer Data Into ChatGPT? — The privacy settings to check before any client work.
- How to Classify Business Data Before Using AI Tools — Sort client data into what AI may and may not see.
- ChatGPT Prompts for Small Business Owners: 50 Tested Examples — Tested prompts for the admin side of a training business.
- AI Subscription Prices Compared for Small Businesses (2026) — Compare ChatGPT, Claude and Gemini plan prices.
- How to Handle Sensitive Customer Conversations Without AI — Client conversations that should stay fully human.
- Gym Retention Software With AI: What It Costs and Delivers — If you work inside a gym, what its retention tools cost.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help pages on Temporary Chat, data controls and Projects in ChatGPT; ChatGPT and ChatGPT Business prices from OpenAI's pricing as of September 2026; ABC Trainerize AI Workout Builder page (inputs, Beta on Grow plan and above, editable before assigning).