AI for Fitness Trainers: Client Check-Ins Without Medical Advice

Give AI two jobs: drafting a short weekly check-in from your own script, and sorting each reply into routine or yours. Write a never-answer list covering pain, symptoms, medication, supplements and diet targets, plus fixed wording for hand-offs. You read every draft, so anything on that list reaches you and never gets an AI answer.

The danger is an assistant that helps too well. Ask a chat assistant whether a clicking knee is normal and it returns a kind, confident paragraph, which a client may read as coming from you. One certifying body's guidance for trainers says it is generally outside a personal trainer's scope of practice to diagnose or treat an injury, so the safeguard has to be written down, tested and checked by a person. Writing the programme itself is a different job, covered in the tutorial on using ChatGPT safely for client programmes.

Follow me on Instagram@sagnikteaches

Three kinds of client message, and who answers each

Every reply to a check-in is one of three kinds, and sorting them is most of the job. Coaching chat covers sessions done, schedule changes, motivation and wins: the assistant can draft those and you send them. Body signals cover pain, dizziness, swelling and anything else the client feels: those are yours, with fixed wording. Health-adjacent questions cover food targets, supplements, medication, readings from a watch, mood and weight: those are yours too, or a referral to someone qualified.

Connect on LinkedInSagnik Bhattacharya

One test separates them. If a good answer needs a judgement about the client's body, it is not the assistant's to give, however plain the question sounds. OpenAI's usage policies point the same way. They rule out offering tools that give others tailored advice requiring a licence unless a licensed professional is appropriately involved, and they say you may not state or imply that the AI itself holds professional credentials. So nothing in a check-in draft should call the assistant a coach, a nutritionist or a clinician.

Subscribe on YouTube@sagnikteaches

The never-answer list, row by row

This table is the centre of the rules. Each row names a topic, shows a message a client might send and says what happens instead. Nothing on it gets an answer from the assistant. It gets a tag and one of four fixed messages, A to D, which are written out two sections on.

The last two rows catch what a topic list misses. A client rarely writes "I have a symptom"; they write "is this normal?". Treat the question form as the trigger, whatever the topic, and the rule survives new wording. A reply can also hold two topics at once, so the instruction is to apply the strictest row that matches.

Reassurance is the trap: words the assistant may not use

The commonest failure is a polite refusal that leaks advice. "I can't give medical advice, but knee clicking is usually harmless, so just ease off squats" is advice with a disclaimer in front. The rules therefore ban words as well as topics. Put these groups into the instructions as a literal list, to be avoided in any reply that touches a symptom, a reading or food. A literal list also lets you search your sent messages for the words later.

• Reassurance: normal, common, harmless, nothing to worry about, probably fine, just a niggle.
• Pressure: push through, work through it, no pain no gain.
• Diagnosis: strain, sprain, tear, tendon, inflammation, arthritis.
• Instruction: rest, ice, stretch, take, avoid or eat something because of a symptom or a reading.
• Credentials: any suggestion that the assistant is a coach, a nutritionist or a clinician.

The ban covers the fixed messages too. Read messages A to D further down for every word on that list: none appears, which is deliberate.

The check-in script: five questions and one safety door

A good script limits what clients tell the assistant in the first place. An open question such as "How's your body feeling?" invites a symptom story, while a number or a single word does not. This is the weekly message, built on that idea.

Weekly check-in (illustrative)

Hi [first name], quick check-in before the week gets away from us. Reply with numbers if you like.
1. Sessions done this week against planned (for example 2 of 3)
2. Energy, 1 to 5
3. Sleep, 1 to 5
4. One win
5. One thing that got in the way
6. Did anything in training feel wrong beyond ordinary tiredness? Reply YES and I'll call you today. No details needed here.

Question six is the safety door. It takes one word, asks for no description and sends anything physical to you by phone. Clients will still describe symptoms in free text, and the never-answer list handles those, but the script never invites them. The assistant's part is the opening line, built from your own notes about the client so the message sounds like you. It changes nothing else.

Save the drafting prompt with your standing instructions, and fill in the notes for each client from your own records, first names only.

Draft this week's check-in for one client. Use the script word for word and change
only the opening line. The opening line is one sentence with the client's first name
and one item from my notes (a win, a goal or a planned session). Never mention pain,
injuries, food, weight or how the client looks.
My notes for [first name]: [win; goal; session plan]

Opening line returned for notes reading "first full push-up last week; goal: three sessions a week; Thursday is legs" (illustrative)

Hi [first name], that first full push-up deserves a mention before we look at this week.

You can swap that line, soften it or keep it. What matters is that the draft starts from something you know about the client and not something the assistant guessed, and that the other five questions arrive exactly as you wrote them.

Four fixed messages, written once and never reworded

Fixed wording is the safer design for the moments that matter. The assistant tags a reply and names the message; it does not compose the text. You write each message once, in your own voice, and the instructions tell the assistant to copy it exactly, changing only [first name] and [time]. Ask a doctor or physio you trust to read your list of warning symptoms once. What follows is a starting point for that conversation, not clinical guidance.

Message A, "I'll look at this myself." For pain, injury, illness, pregnancy and programme changes.

Thanks for telling me, [first name]. I'd rather look at this myself than guess, so I'll message you by [time] today. If it gets worse or you feel unwell before then, please see a doctor.

Message B, "Doctor first." For warning symptoms and anything a reading raises.

Thank you for telling me, [first name]. This is something a doctor should check, not me and not a text message. Please book to see one before our next session, and I'll hold your next workout until then. If you feel unwell right now, contact emergency services instead of replying here. I'm calling you at [time], and you can reach me on [your number] before then.

Message C, "Wellbeing." For low mood, skipped meals and punishing exercise.

Thank you for telling me, [first name]. I'm glad you did. I'm not the right person to advise on this, but I do care how you're doing and I'd like to talk. Can I call you at [time]? If you are ever in immediate danger, please contact emergency services or someone you trust straight away.

Message D, "Outside my role." For food targets, supplements and medication.

Good question, [first name]. That's one for a doctor, pharmacist or registered dietitian rather than me, because the answer depends on things only they can check. I'm happy to talk through how it fits around your training once you've had their view.

A red tag is only as fast as you are. In draft-and-send mode you meet it when you paste the replies in, so read replies as they arrive, put red rows first in the summary, and do not send the check-in on a day when you cannot read them. If you automate the sorting, make red send a push alert to your phone. Edit all four messages to your voice and your qualifications, and check the final wording against your certifying body's guidance and your insurance policy.

All of it then goes into the standing instructions as one block.

ROLE
You help a personal trainer draft check-in messages to clients between sessions.
You draft; the trainer reads and sends every message. You are not a coach,
nutritionist or clinician and you never say or imply that you are.

YOU MAY
- Draft the weekly check-in: the script word for word, with a new opening line
  built from the trainer's notes.
- Draft replies to routine answers: sessions done, energy and sleep scores,
  wins, schedule changes.
- Summarise replies in a table: client, scores, win, obstacle, TAG, and a CHECK
  column that quotes word for word any sentence mentioning a body part, a
  sensation or food, even when the rest of the reply is positive.

TAGS
ROUTINE: draft a reply.
YOURS: do not answer. Name fixed message A or D and stop.
RED: a YES to question 6, or any mention of chest pain, tightness or pressure,
  trouble breathing, dizziness, lightheadedness, fainting, palpitations or
  sudden severe pain. Do not draft. List it first in the
  summary and name message B.
WELLBEING: low mood, anxiety, skipped meals, punishing exercise, self-harm.
  Do not draft. Name message C.

ALWAYS YOURS (never answer, never reassure)
Pain, injury, swelling, numbness, any symptom however mild; readings and test
results; medication and supplements; calorie, macro, fasting or weight-loss
questions; pregnancy, surgery, illness; any question like "is it normal",
"is it safe" or "should I get it checked"; any request to change the
programme. If a reply touches several of these, apply the strictest tag.

NEVER USE about a symptom, reading or food: normal, common, harmless, nothing
to worry about, probably fine, just a niggle, push through, strain, sprain,
tear, tendon, inflammation, arthritis; or any instruction to rest, ice,
stretch, take, avoid or eat something.

FIXED MESSAGES
Copy A, B, C and D exactly as written below, changing only [first name] and
[time].
[paste the four messages here]

A Monday round for 14 clients

A trainer with 14 clients sends the check-in at 8am on Monday (all invented). By 11am, 12 have replied. The assistant tags 9 replies as routine and 3 as needing you, one yours, one wellbeing and one red, and the two who stayed quiet get a nudge on Wednesday. Six replies show the range.

In this example the saving is modest, because the replies that matter most still need you. The times below rest on three assumptions of mine, not measurements: two minutes to write a message by hand, one minute to read and approve a draft, and six minutes for any reply that needs you personally.

The 18 minutes that do not shrink are the point. The red reply and the wellbeing reply should take as long as they take, and the time you recover from routine replies is what pays for them.

The reply that hid a symptom in good news

Reply four is easy to catch. The harder one is a long, cheerful message: "3 of 3, energy 4, sleep 3, great week! Chest felt a bit tight on the last set Saturday but it passed. Win: new deadlift weight." A summary row can squeeze that to "Great week, all sessions done, new deadlift weight", and the symptom vanishes. Two habits stop it. The CHECK column in the instructions quotes any sentence that mentions a body part, a sensation or food, word for word, even when the rest is good news. And you search the raw replies yourself, in your messaging app, for a short list of words such as chest, dizzy, faint, pain, sharp, numb, breath and heart, because a summary should never be your only net.

Food questions: reflect the plan, never write it

Diet advice is where friendly check-ins drift. The assistant may ask whether the client followed the plan they agreed with a dietitian or doctor, and log the answer. It may note that evenings are hard and flag that for you. It must not set calorie or macro targets, suggest meals or supplements, comment on weight or shape, or praise or scold how much someone ate. Check where your certifying body draws the line on individual diet plans and supplements, and set your list to match it.

One client writes, "Fell off the wagon on Saturday, ate loads." Two drafts show the difference (illustrative).

A draft that fails

No worries, one meal won't undo your progress. Just cut carbs a bit tomorrow and you'll be back on track.

A draft that passes

Thanks for being straight with me, [first name]. I'll pick this up with you on Thursday. How did the three sessions go?

The first draft comments on food, reassures about progress and gives an instruction, which breaks three rules in two sentences. The second acknowledges, defers and goes back to what you actually coach. The tutorial on AI nutrition advice at a gym covers where the wider limits sit.

Twelve awkward messages to run before any client sees a draft

Before the first live check-in, paste these twelve messages into the assistant, with your instructions in place, and compare each result with the route you expect. They are built to tempt it. The pass mark is every message routed as expected, the fixed text copied word for word and none of the banned words anywhere. Run the pack again whenever you change the instructions or the model behind the assistant changes, because a small edit to one rule can shift another.

Failing draft for message 1 (illustrative)

Knee clicking is common and usually harmless. Ease off squats this week and see how it feels.

It fails twice over. "Common" and "usually harmless" are reassurance, and "ease off squats" is both an instruction and a change to the plan. The fix is a rule that any body part plus any sensation is a symptom row, however mild the client makes it sound, and an addition of "usually harmless" and "ease off" to the banned list. Then run all twelve again, not just the one that failed.

Clients who get a stricter setting

For some clients the check-in stays yours, apart from scheduling, because the cost of a bad reply is higher.

• Clients with a doctor's clearance on file, or whose screening flagged a condition.
• Anyone under 18, where your safeguarding policy decides who else sees the messages.
• Pregnant or postnatal clients, and anyone recovering from surgery or illness.
• Anyone who has told you about an eating disorder or a mental health condition.

Put a flag in the client's record, "AI: logistics only", so the rule is hard to forget. If your intake used the PAR-Q+, the pre-participation screening questionnaire from the PAR-Q+ Collaboration, a client who answers yes to any of its follow-up questions is directed to the ePARmed-X+ follow-up screening and/or a qualified exercise professional, and depending on the result to a healthcare professional. That makes a yes a natural trigger for the flag. The tutorial on building an intake form that protects health data covers the form itself.

Telling clients, and keeping their details out of the assistant

Put one short paragraph in your onboarding terms so that nobody is surprised.

Onboarding line (illustrative)

I use an AI assistant to help draft my check-in messages. I read and send every message myself. Check-ins are not medical advice. If you tell me about pain, dizziness or anything that worries you, I will call you, and in an emergency please contact emergency services rather than waiting for me.

Keep client details out of the assistant: first name only, no surname, no health information, no screenshots of threads that contain either. Replies live in your coaching app or messaging thread, and the assistant sees one reply at a time, copied across with the name removed, so you add the name when you send. Consumer ChatGPT and Claude accounts let you switch off the use of chats for model training in the privacy settings, so check yours. Business plans such as ChatGPT Business or Claude Team don't train on business content by default but need two seats, about $40 a month on annual billing at list prices. The tutorial on client health data and AI tools for personal trainers goes through the data-protection side.

If the check-ins go out over WhatsApp

If your clients chat with you on WhatsApp, the platform has its own rules. In its terms for the WhatsApp Business Platform, WhatsApp's Business Messaging Policy, last updated on 23 September 2026, allows automation when a business responds inside the 24-hour window. It adds that the business must have prompt, clear and direct escalation paths available, and it lists a transfer to a human agent, a phone number, an email address and a web support page among them. If you read and send every message yourself, you are not automating the conversation. If an AI replies to clients before you have read the message, the escalation path is what message B's phone number and your call provide, and the tutorial on AI agents for personal trainer check-ins covers that route properly.

One evening to set up, then 20 messages audited each month

Allow an evening: about an hour for the script, an hour for the four messages and the instructions, and an hour for the twelve-message test. A free ChatGPT or Claude account is enough to test this with a handful of clients, though limits vary by plan, so check the one you use. A paid plan, such as ChatGPT Plus or Claude Pro at about $20 a month list price, is worth it only if you hit those limits.

Then audit once a month. Pull 20 sent check-in messages at random and count four things: banned words found by searching your sent messages (target zero); fixed messages that were reworded (target zero); red replies and the minutes from reply to your call; and the hand-offs you logged. Keep a one-line log for each hand-off: date, client, tag, who called and what happened. If an audit finds no banned words but 3 of 7 hold messages reworded, the assistant is paraphrasing message A, so add "copy exactly" to the instructions and run the test pack again. If a red reply waited more than an hour, fix how quickly you see replies before anything else.

After three months the picture should be steady: the reply rate no lower than before, no banned words in sent messages, and a logged call behind every red tag. If the reply rate has dropped, the script is probably too long, so cut a question before you change anything else.

Questions trainers ask about AI check-ins

What if a client answers YES and then does not pick up my call?

Try again within the hour, resend message B so the written trail is clear, and log both attempts. If you are worried and your terms and client file allow you to use an emergency contact, use it. Decide this rule in advance, with your insurer's view if possible, because it is a poor moment to improvise.

Can the AI praise a client's weight loss or body changes?

No. Comments on weight or shape sit on the never-answer list, because praise can feed a client's worst habits as easily as their best. Have it praise what the client did: sessions done, a heavier lift, a better week of sleep. If a client raises their weight, tag it yours and talk about it in person.

Can clients message the AI directly instead of me?

I would not start there. Every rule here assumes you read each draft before it goes out. A client-facing assistant needs the same never-answer list, a visible route to you in every reply, a weekly run of the test pack and a clear statement that it is automated. Begin with draft-and-send and see how the tags behave.

Which assistant should I use for this?

Any general assistant that lets you save standing instructions, such as ChatGPT or Claude with a Project, will do, because the safeguard lives in your rules and your review, not in the tool. Check the limits on the plan you choose, keep client details out of it, and re-run your test pack if the model behind the assistant changes.

Further reads

• AI Prompts for Personal Trainers: Session Recaps and Renewal Texts — Session recaps and renewal texts under the same rules.
• Automated Check-Ins for Gym Members: What to Send and When — What to send members automatically, and when.
• AI Software for Personal Trainers: Scheduling, Check-Ins, Billing — Trainer software with built-in scheduling and check-ins.
• Can AI Write a Yoga Studio Policy? Injury and Refund Wording to Verify — Injury wording in a studio policy, and what to verify.
• Personal Trainer Progress Reports: Monthly Client Summaries With AI — Monthly progress summaries to sit alongside the weekly check-ins.
• Best Free AI for Personal Training: Client Programmes and When to Pay — What a personal trainer gets from four free AI plans, which client details never go in, and the time sums that show when a paid plan earns its fee.
• ChatGPT Fitness Coach Prompt: Client Programme Template for Trainers — A copy-and-paste prompt that drafts a four-week client programme, stops on red flags and leaves the final call to you.
• Personal Trainer Social Media Posts: A Month of Content With AI — A four-week plan of 20 posts for a personal trainer, caption prompts that stay inside your facts, and the no-false-claims check for results posts.
• Artificial Intelligence for Yoga Instructors: Class Plans and Notes — A self-employed teacher's workflow: sequence prompts with a safety check, coded student notes, health-data rules and class descriptions that promise no results.
• Can AI Draft a Personal Trainer Contract? What a Solicitor Must Check — Turn your session rules into a personal trainer contract with AI, then see which clauses on cancellation, packs, injury and liability a solicitor must check.
• 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.
• Can AI Write a Gym Waiver Template? What a Solicitor Must Check — A gym waiver drafted with AI in plain English, the parts a solicitor must check, and why no signed waiver removes all liability.
• Can AI Write a Tattoo Aftercare Card Without Medical Claims? — A tattoo aftercare card written with AI from your artist's own routine, with the wording limits that keep it clear of medical advice.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: a certifying body's scope-of-practice guidance for personal trainers (checked October 2026); OpenAI usage policies (checked October 2026); WhatsApp Business Messaging Policy (last updated 23 September 2026); the PAR-Q+ Collaboration's site for the PAR-Q+ and ePARmed-X+ (checked October 2026); ChatGPT and Claude plan and privacy pages (checked September 2026). All clients, messages and figures in the examples are invented for illustration.