GDPR for Personal Trainers: A Checklist for Client Health Data and AI

If you have clients in the EU, the GDPR treats a trainer's health notes as special-category data. You will usually need explicit consent, a privacy notice, a retention schedule, a written contract with every app or AI provider that holds the data, and a breach routine. Keep identifiable health details out of consumer AI chats.

The same duties apply to a sole trader and to a chain, though a small studio can keep the paperwork short. What catches small studios out is rarely the law itself. It is a questionnaire sitting in a shared folder, a progress photo in a WhatsApp group, or a client's injury history pasted into a free chatbot. None of this is legal advice, and the last group in the checklist lists what to put to a data-protection adviser.

Follow me on Instagram@sagnikteaches

What counts as health data in a trainer's file

Article 4 of the GDPR defines data concerning health as personal data about a person's physical or mental health, including the provision of health care services, that reveals information about their health status. Article 9 then prohibits processing it unless a listed condition applies. For a trainer that covers far more than a medical form.

Connect on LinkedInSagnik Bhattacharya

• The activity-readiness questionnaire and its free-text answers: injuries, surgery, conditions, medication, pregnancy, blood pressure.
• Session notes that record why a programme looks the way it does, such as "lower-back pain, avoid loaded forward bending".
• Body measurements, weight and progress photos, which can reveal health status.
• Food diaries and wearable data such as heart rate and sleep.
• Messages in which a client mentions a diagnosis, a mood or a mental-health concern.

Treat the whole client file as special-category unless your adviser says part of it is not. The practical effect is two extra layers. You need an Article 9 condition on top of an ordinary lawful basis, and the bar for security, and for sharing the data with anyone including an AI tool, goes up.

Subscribe on YouTube@sagnikteaches

Map what you hold before anything else: a solo trainer's inventory

Take a solo trainer with 40 active clients and 65 former ones, so 105 client files in all. A first pass at where the data sits might look like this (illustrative).

That is five problems after an hour's work: an open link, an unchecked contract, photos mixed with personal ones, chats with no deletion rule and a chatbot account with an unchecked training setting. There is still no retention rule overall. Suppose the trainer then sets a rule of keeping health records for 12 months after the last session, and 48 of the 65 former clients are past that point. Forty-eight questionnaires and 48 sets of notes can go, which leaves 57 files: the 40 active clients and 17 recent leavers.

The same table is the start of a record of processing. Article 30(5) exempts organisations with fewer than 250 staff from keeping one unless the processing is likely to risk people's rights, is not occasional, or includes special-category data. Regular health data fits the last two, so a trainer should expect to keep a short written record covering purposes, who receives the data, how long it is kept and how it is secured. Your adviser can confirm.

Before the first form: lawful basis, consent and the privacy notice

These items decide whether you may collect health information at all. Each one has a reason and a quick way to check it.

1. Cut the intake form down to what changes the programme. Article 5 asks for data that is adequate, relevant and limited to what is necessary. Check: beside each question write one sentence on how the answer changes what you prescribe, and delete any question that gets no sentence. "Do you have any mental health conditions?" can become "Has your doctor advised you to avoid anything when exercising?" Building the form itself is covered in personal trainer intake forms and health data.
2. Ask for explicit consent to the health questions, on a separate tick box. Consent has to be freely given, specific, informed and unambiguous, shown by a clear affirmative action, and you must be able to demonstrate it. Withdrawing must be as easy as giving it. The health-care condition in Article 9 is tied to professionals under a secrecy obligation, so do not assume it covers a fitness business. Check: show the form to a friend outside the trade and ask what they just agreed to. If they cannot say, rewrite it.
3. Keep unrelated asks out of the consent. Article 7 tells you to take account of whether a service is made conditional on consent to processing it does not need. Marketing emails and testimonial permission belong in their own boxes, and training should not depend on them.
4. Give the privacy notice before the form, not after. Article 13 lists what to tell people when you collect their data: who you are, why you use it and on what basis, who receives it, any transfers abroad, how long you keep it, their rights including withdrawing consent, and the right to complain to the regulator. Check: it names every tool that will see client data, AI tools included.
5. Treat progress photos as their own yes. A photo can reveal weight, medical conditions or injuries. Ask separately, say where the photos are kept and for how long, and never use them in marketing under the training consent.
6. Handle under-18s with your adviser. The age at which a child can consent for themselves differs between countries. Decide how a parent or guardian takes part before you accept a younger client.
7. Collect emergency contacts sparingly. They are someone else's data. Take a name and a number, and tell the client to let that person know.

The two pieces of wording below are illustrative, with square brackets where your details go. Your adviser should check both before you use them.

[ ] I agree that [business name] may collect and use the health information
I give on this form (injuries, medical conditions, medication, pregnancy and
similar) to plan and adjust my training safely. I can withdraw this consent
at any time by telling [contact]. [Business name] will then stop using this
information and delete it unless the law requires it to be kept.

[ ] Optional: I agree to progress photos being taken and stored for my own
programme only. They will not be shared or used for marketing.

Your information. I am [trainer name], trading as [business name]. I collect
your contact details, payment records and the health and fitness information
you give me, so that I can plan and run your training. I use health information
only with your explicit consent. I share it only with [coaching app], [cloud
storage provider] and [payment provider], who process it for me under written
contracts. I never put your name or contact details into an AI tool, and I
remove anything that identifies you before I use one. I keep client records for
[period] after your last session, then delete them. You can ask to see, correct
or delete your information, or withdraw consent, by contacting [contact]. You
can also complain to your data-protection regulator.

While you hold it: storage, access and every app that touches the file

Once consent is in, the risk moves to where the data lives. Article 5 asks for processing that keeps data secure, and a trainer's version of that is mostly habits.

1. Keep client files in one place, behind a password manager and two-step sign-in. Check: list every place a questionnaire could be hiding, including email attachments, the downloads folder, phone photos and WhatsApp media, then delete the stray copies. Turn on device encryption and a short automatic screen lock on the phone you carry to sessions, because a lost phone is the most likely breach.
2. Stop link sharing on health folders. A folder shared with "anyone with the link" is open to anyone who finds the link. Share with named people only. Check: open the folder's sharing settings and read the list aloud.
3. Sign a processor contract with every app that stores or sees client data. That means the coaching app, cloud drive, booking tool, email platform and payment provider. A processor is anyone handling data on your behalf, and Article 28 says the arrangement must be governed by a contract requiring them to act only on your documented instructions, keep staff bound to confidentiality, apply security measures, get your authorisation before bringing in sub-processors, and delete or return the data when the service ends. Check: for each vendor, find its data processing addendum, confirm it applies to your plan, and note the link and date in your record.
4. Ask where the data is processed. If a provider handles data outside the EU, the contract needs a transfer mechanism. OpenAI's addendum refers to standard contractual clauses for transfers, and Anthropic's addendum includes them, with its own breach-notification promise of 48 hours. Check: read the transfer clause and the sub-processor list, and put any doubt to your adviser.
5. Keep progress photos out of the camera roll. Use the coaching app's storage if it has one, or a locked folder that does not sync to a family account, and delete the photos when the programme ends.
6. Treat chat apps as a leak risk, not a filing system. A chat becomes a shadow client file within weeks. Move anything you need into the client record, and use the app's disappearing-messages option for client chats if you keep them. Never send a programme that carries health notes to a group chat.
7. Lock up paper. If questionnaires arrive on paper, scan them into the client file and shred the sheet, or keep them in a locked drawer with a deletion date written on the front.

A quick before and after shows the sharing fix. Before: the questionnaire folder is shared by link so that new clients can upload forms, and the link was once pasted into a social post. After: clients fill in a form that writes to the folder, the folder is shared with the trainer only, and the old link is switched off. The same questionnaires exist in both versions. What changed is who can reach them.

For a wider read on the habits that keep customer data out of the wrong places, see keeping customer data private when your team uses AI.

AI tools: what may go in, and the settings to check first

The AI part of the checklist has one rule and a handful of checks. The rule: no identifiable health data goes into a consumer AI account. Identifiable means a name, birth date, phone number, email address, photo or voice note, and any combination of details that points to one person. A consumer chat is not a processor contract, and what you type may be used to train the model unless a setting says otherwise.

The vendors describe their own defaults as follows (checked this month; setting labels change often, so look for the equivalent switch if yours reads differently).

The detail sits on the vendors' own pages: OpenAI's Data Controls FAQ and business data page, and Anthropic's privacy centre article on model training. Two cautions. First, a vendor's business terms only help if your plan is covered by them, so confirm that before relying on the addendum. Second, business plans have seat minimums: Claude's pricing page lists Team for 2 to 150 people, at $20 per Standard seat a month billed annually or $25 billed monthly, and ChatGPT Business is available from 2 users. A one-person trainer would pay for two seats, which is the price of the safer route.

Whether an AI provider is GDPR-compliant is a question with more layers than a table can hold; are ChatGPT, Claude, Gemini and Copilot GDPR-compliant goes through it, and zero data retention explains what the strictest retention promise does and does not mean.

Five checks before any client detail reaches an AI tool

1. Switch off training on every personal account you have ever used for work. Do it once, write the date in your record and look again after any big app update. Check: take a screenshot of the setting for your file.
2. Anonymise before you paste. Remove names, birth dates, contact details, dates that identify an event, medication names and anything unusual enough to point to one person. Keep the clinical detail the task needs. The key linking "C-014" to a real client stays in the client file, never in the AI account.
3. Never upload photos, voice notes, scanned questionnaires or raw wearable exports. These carry identity on their face, in their metadata or in the voice, and anonymising them properly is harder than it looks.
4. Use a business plan with a contract if AI will ever touch identifiable data. That is the only route that gives you a processor agreement to point to.
5. Treat every output as a draft for you to review. An AI programme is not medical advice, and anything touching a condition is a question for the client's clinician. The checking routine in using ChatGPT safely for client programmes goes into this.

Here is the anonymising step as a before and after (illustrative, with placeholders). The first version must not be pasted anywhere.

Before: [first name] [surname], born [date], phone [number]. Eight weeks after
knee ligament surgery, takes [medication], cleared by her physio for light
loading, nervous about squats, wants to lose weight before a wedding in June.

After: Client C-014, adult in their 30s, eight weeks after knee ligament
surgery, cleared by their physio for light loading, nervous about squatting,
goal is general conditioning over 12 weeks. No other limitations reported.

Pseudonymised text like the second version is still personal data. Article 4 describes pseudonymisation as processing data so that it can no longer be attributed to a person without additional information that is kept separately. That lowers the risk, and it leaves your duties in place, because you hold the key.

Draft a 12-week lower-body conditioning outline for client C-014 (details
above). Three sessions a week, 45 minutes each, with progressions every four
weeks. Flag any exercise a physio should approve first. Give no medical advice
and no return-to-sport dates. End with five questions to ask the physio.

A typical reply (illustrative) would give a week-by-week outline, a list of flagged movements such as deep loaded knee bending, jumping and sharp changes of direction, and five questions for the physio. Expect to fix at least one thing. The model may slip in a line like "return to jogging by week 10", which the prompt forbids and which only the physio can decide. Delete it, and keep the flag list as your agenda for the clinician.

One more habit: vendors change defaults. In June 2026, for example, SimplePractice's Note Taker began opting new users in by default to keeping de-identified transcripts. A tool you configured safely last year can behave differently now, so re-read the privacy settings of anything that holds client notes whenever the vendor announces changes.

When clients ask, leave or withdraw: access, deletion and retention

1. Practise one access request before a real one arrives. Article 12 says to respond without undue delay and within one month of receiving the request, extendable by two further months where necessary, with the client told about the extension and the reasons inside the first month. Responses are generally free of charge. Say a request lands on 3 March. The answer is due around 3 April, and a justified extension, notified by then, could run to about 3 June. Check: pick one former client and try to list every place their data sits, chats and photos included, in 30 minutes. The inventory table earlier is what makes that possible.
2. Know what you do when consent is withdrawn. You stop using the health information, delete it unless another reason or duty justifies keeping it, and confirm in writing. Whether anything justifies keeping it, such as an insurer's expectation, is a question for your adviser. Check: write the procedure as one paragraph and keep it in the same folder as the consent form.
3. Set a retention schedule and follow it. The GDPR gives no fixed period. Article 5 says data may be kept in identifiable form for no longer than necessary, and your record of processing should state the erasure time limit for each category where possible. The table gives a starting point. The periods are illustrative placeholders to agree with your insurer and adviser, not legal limits.

4. Run an exit routine when a client leaves. Hand over their programme if they ask, note the deletion date from your schedule, remove them from the coaching app and shared folders, take them off any marketing list, and delete the progress photos. Check: after the date passes, search each place on the inventory for their name.

If something goes wrong: a breach drill you can run in ten minutes

Article 4 defines a personal data breach as a security breach that leads to the accidental or unlawful destruction, loss, alteration or unauthorised disclosure of, or access to, personal data. A phone left on a train, a programme sent to the wrong client and a hacked email account all count. So might an identifiable client note pasted into a personal AI chat.

1. Contain it. Change passwords, remote-wipe the phone, ask a wrong recipient to delete the message, delete the AI chat and check its training setting.
2. Write down when you became aware. Article 33 says to notify the regulator not later than 72 hours after becoming aware, unless the breach is unlikely to result in a risk to people, and to document every breach either way.
3. Assess the risk to the clients. Health data raises it. Encryption lowers it.
4. Tell the clients if the risk is high. Article 34 requires you to tell them, unless the data was unintelligible to anyone not authorised to see it (for example because the device was encrypted) or later steps mean the high risk is no longer likely.
5. Record what happened and what you changed.

A worked timeline (illustrative): a phone is left on a train at 17:00 on Friday and the trainer realises at 09:00 on Saturday. The 72 hours run from awareness, so they expire around 09:00 on Tuesday. If the phone was encrypted, locked and remotely wiped, the trainer may be able to show the breach is unlikely to risk clients, but the incident still goes in the log. Your adviser should confirm the call.

Seven questions to take to a data-protection adviser

A first meeting goes further if you arrive with the inventory, the consent form and the privacy notice in hand. These are the questions that checklist cannot answer for you, because the answers depend on where you trade and what you offer.

1. Which Article 9 condition, and which ordinary lawful basis, should I record for each kind of data I hold?
2. Do I need a record of processing, and does mine cover enough?
3. Is a data protection impact assessment needed? Article 35 points to large-scale processing of special-category data, so ask whether a solo practice counts.
4. Must I register with my regulator, or pay a fee, where I trade? This varies from country to country.
5. Which of my tools are processors, and do their agreements cover my plans and any transfers abroad?
6. How long should I keep each record, given my insurance and tax rules?
7. How should I handle clients under 18, and clients who ask for deletion halfway through a programme?

Keep the evidence in one folder so you can show it to an adviser, an insurer or a client who asks how you look after their information: the consent form and privacy notice with version dates, the inventory or record of processing, each vendor's addendum with the date you checked it, the retention schedule, screenshots of the AI training settings, and the breach log, even if it is empty.

Trainer questions about client data and AI

Is a client's training history health data under the GDPR?

It can be. The GDPR defines data concerning health as personal data about a person's physical or mental health that reveals information about their health status. Injuries, conditions, medication, pregnancy and the reasons behind a programme all qualify, and measurements or photos may reveal health status too. Treat the whole client file as special-category data unless your adviser says a part of it is not.

Do I need consent to store a client's health questionnaire?

Special-category data is prohibited unless an Article 9 condition applies, and explicit consent is the one most trainers rely on, because the health-care condition is tied to professionals bound by a secrecy obligation. Use a separate, specific tick box, record the date and version, and make withdrawal as easy as giving consent. Ask an adviser to confirm what fits your services.

Can I paste client notes into ChatGPT or Claude?

Not identifiable ones into a personal account. Personal ChatGPT workspaces have the Improve the model for everyone setting on by default, and Claude's consumer plans have a Model Improvement setting. Business plans from both vendors are not trained on by default. Remove names, birth dates, contact details and photos first, and remember that pseudonymised notes are still personal data.

How long should a personal trainer keep client records?

The GDPR sets no fixed period. It says data should be kept in identifiable form for no longer than necessary. Set a written schedule for each type of record, usually shorter for health notes and photos than for invoices, and ask your insurer how long it expects session records to be kept. Tax rules may require invoices to be kept longer than health notes.

How fast must I answer a client who asks for their data?

Without undue delay and within one month of receiving the request. You can extend by two further months where necessary, but you must tell the client about the extension and the reasons within the first month. Requests are generally free of charge. Knowing where each client's data sits, including chats and photos, is what makes the deadline realistic.

Further reads

• Can AI Draft a Personal Trainer Contract? What a Solicitor Must Check — What a solicitor must check in an AI-drafted client contract.
• AI for Fitness Trainers: Client Check-Ins Without Medical Advice — Client check-ins that stay clear of medical advice.
• What to Check in an AI Tool's Privacy Policy and Terms — A method for reading any AI tool's privacy terms.
• Can AI Write a Gym Waiver Template? What a Solicitor Must Check — The waiver side of the same paperwork.
• Best Free AI for Personal Training: Client Programmes and When to Pay — Where the free AI plans stop for client work.
• GDPR for Photographers: Client Galleries, Contracts and AI Tools — The same duties for another client-facing small business.
• AI Agent for Personal Trainers: Check-Ins and Where It Must Stop — Where an AI agent for client check-ins must stop.
• 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 Prompts for Personal Trainers: Session Recaps and Renewal Texts — Ten copy-ready prompts with sample outputs for trainer session recaps, check-in messages and renewal texts, plus rules on what AI must not say.
• ChatGPT Apps for Fitness Studios: Seven Jobs and Their Limits — Seven fitness-studio jobs for ChatGPT's connected apps, the permission to set for each, and the steps that stay with you.
• 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 Progress Reports: Monthly Client Summaries With AI — A worked month for one client: measurement table in, honest report out, with a three-step check so the AI never adds results you didn't record.
• 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 a Personal Trainer Edit Testimonials With AI? What Is Allowed — What AI may and may not do to a client's testimonial: consent, result claims, light edits and a four-minute check before you post.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the text of the GDPR (Articles 4, 5, 7, 9, 12, 13, 28, 30, 33, 34 and 35); OpenAI's Data Controls FAQ, business data page and Data Processing Addendum; Anthropic's privacy centre articles on model training and data retention, and its Data Processing Addendum; Claude and ChatGPT pricing pages.