EU AI Act for fitness apps and coaches: Article 50

Does the EU AI Act apply to my fitness app or coaching business?

If you offer an AI chatbot or AI coach that people in the EU talk to, yes: since 2 August 2026, Article 50 of the AI Act requires the provider to tell users they are dealing with AI, at the latest at the first interaction. Generative systems must also mark their output in a machine-readable way, and systems already on the market before 2 August 2026 have until 2 December 2026 for that part. A coach who only uses ChatGPT to draft plans is usually a deployer, with far fewer duties.

Updated 28 October 2026

This guide is written for three people: the coach who drafts programmes with a general chatbot, the fitness app that ships its own AI coach, and the white-label partner that sells someone else's AI under its own brand. Every legal point below links to the text of Regulation (EU) 2024/1689 on EUR-Lex, the European Commission's AI Act Service Desk or the Commission's own guidance. It is not legal advice, and where the law leaves a question open, this article says so instead of guessing.

Three fitness roles at a glance

The roles come from the definitions in Article 3: a provider develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark; a deployer uses an AI system under its authority, except for personal non-professional use.

What Article 50 actually requires

Article 50 contains four separate duties. Only the first two land on app makers; the other two land on whoever uses the system.

• Interactive AI (50(1)), provider duty. A system meant to interact directly with people must be designed so they know they are talking to AI, unless that is obvious to a reasonably well-informed, observant person in the context.
• Generated content (50(2)), provider duty. Systems that generate synthetic audio, image, video or text must mark outputs in a machine-readable format so they are detectable as AI-generated. Tools that only do standard editing, such as grammar fixes, are exempt to that extent.
• Emotion recognition and biometric categorisation (50(3)), deployer duty. If you use such a system, you must inform the people exposed to it and process their data under the GDPR.
• Deep fakes and public-interest text (50(4)), deployer duty. Disclose deep fakes. Disclose AI-generated text published to inform the public on matters of public interest, unless a human reviewed it and a person holds editorial responsibility.

Article 50(5) adds the how: the information must be clear and distinguishable, given at the latest at the first interaction or exposure, and meet accessibility requirements.

The dates and the penalty ceiling

• 20 July 2026: the Commission published its guidelines on the Article 50 transparency obligations (full text).
• 27 July 2026: the Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force, as the Commission's AI Act page and the AI Act Explorer confirm. It pushed high-risk rules for Annex III areas to 2 December 2027. It did not move Article 50.
• 2 August 2026: the AI Act applies in general under Article 113, Article 50 included, also for systems that were already running.
• 2 December 2026: the one transition the Omnibus added, in Article 111(4): providers of generative systems placed on the market before 2 August 2026 must comply with the marking duty of Article 50(2) by this date. The chatbot disclosure has no such grace period.

The Commission's guidelines add a useful detail: outputs generated before 2 August 2026 do not need to be marked retroactively. Old AI-written plans in a client's history are not the problem; new output is.

Fines are set by each member state, but Article 99(4) caps breaches of Article 50 at 15 million euros or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups, Article 99(6) flips that to whichever is lower, and member states must take the economic viability of small businesses into account.

If you are a coach who drafts plans with ChatGPT

In practice this is most coaches reading this. You use a general chatbot in your business, so you are a deployer of that system. The disclosure duty of Article 50(1) and the marking duty of 50(2) belong to the company that provides the chatbot, not to you.

Two things still apply.

First, the public-interest text rule. A training plan sent to one client is not "published" text: the guidelines name private, interpersonal correspondence for professional purposes as an example of text that is not published. Your blog is different. Among the Commission's own examples of text that must be labelled is an AI-edited lifestyle article comparing diets for a particular disease. So if you publish AI-written articles on nutrition and health, either a qualified person reviews the substance and you hold editorial responsibility, or you label them. A quick spell-check does not count as review, according to the guidelines.

Second, AI literacy. Article 4 asks providers and deployers to take measures to support the AI literacy of staff who operate AI on their behalf. For a solo coach, that mostly means knowing where the tool is weak: made-up numbers, ignored injuries, plans that do not match the intake.

The AI Act does not force you to tell a client that ChatGPT helped write their plan. Telling them anyway is simply good practice, and the guidelines point out that for a service sold as AI-driven, such as a virtual coaching service, consumer law may treat the AI part as a main characteristic to disclose before the contract. If you want better drafts, our ChatGPT prompts for fitness coaches cover the prompting side.

If your app has its own AI coach

You are the provider, and the scope rule in Article 2 applies to providers placing systems on the EU market wherever they are based, and to non-EU providers whose output is used in the EU. A US app with German users is in scope.

The guidelines are concrete about what works. Good examples: a chatbot that opens the conversation by saying it is AI, a plain banner such as "You are interacting with an AI system", a persistent AI badge near the input field. Not enough on their own:

• a line in the terms and conditions or the privacy policy
• a vague label such as "assistant", or a human name and photo that could mislead
• a site-wide note that "services on this website use AI"
• a technical statement like "powered by an LLM" that does not say the user is talking to a machine

Health advice is one of the contexts where the guidelines expect more than one notice: periodic reminders are likely necessary where users could be misled. The guidelines also expect the system to disclose that it is AI whenever a user asks about its nature.

Then there is marking. If your AI coach writes plans, meal suggestions or images, those are generated text and images under Article 50(2). If the feature was live before 2 August 2026, the deadline is 2 December 2026; a feature launched later had to be marked from day one.

If you sell a white-label AI coach under your brand

This is where the honest answer is "it depends". The provider definition in Article 3 covers anyone who has an AI system developed and places it on the market under its own name or trademark. The guidelines add that a company which takes an existing generative system, modifies it and puts it into service under its own name becomes the provider of the new system. The explicit rebranding rule in Article 25 is written for high-risk systems, and the guidelines do not discuss white-label resale by name.

What follows in practice: put it in the contract. Agree who builds the AI disclosure into the interface, who handles machine-readable marking, and whether you may rename the AI persona. Renaming it to a human first name with a stock photo is exactly the kind of human-like representation the guidelines warn can mislead. Then ask a lawyer to confirm your role. If you are weighing a branded app in general, our guide to white-label fitness apps for trainers covers the business side.

Where fitness apps touch the riskier edges

Article 50(3) applies to emotion recognition and biometric categorisation. Both are defined in Article 3 as working on biometric data. A mood score the client types in is not that. A feature that reads stress or emotion from a face or voice, or that sorts people into categories from body photos, may be, and then the deployer must inform the people exposed and follow the GDPR. Check the definitions before you build, and read our piece on AI fitness app privacy for the data side.

One more line people mix up: an AI disclosure is not a medical disclaimer. "You are chatting with AI" does not tell a client with an injury, a pregnancy or an eating disorder history to see a professional first, and a medical disclaimer does not satisfy Article 50. You need both. Our article on when to refer a client helps with the second.

What Trainera does here

TRAI, the AI assistant inside Trainera, always tells the user that it is AI. Garmin data that clients sync to Trainera is never used by TRAI or any other Trainera AI. You can see how TRAI works for coaches and clients on the AI fitness coach page, and our sibling article on what an AI personal trainer can and cannot do covers the coaching side.

Your Article 50 checklist for this week

1. Write down your role for every AI tool you use or sell: provider, deployer, or unclear.
2. If you provide a chatbot, open it as a new user and check that the very first screen says it is AI, in every language you support.
3. Remove human names or photos from AI personas unless the AI label sits right next to them.
4. List every feature that generates text, images, audio or video, and note the date it went live. Anything live before 2 August 2026 needs machine-readable marking by 2 December 2026.
5. For your blog, decide per article: human review with a named responsible person, or an AI label.
6. If you resell AI under your brand, get the split of Article 50 duties into the contract.
7. Keep your medical disclaimer and your AI disclosure as two separate, visible messages.

Want an AI assistant that says it is AI from the first message? See how TRAI works in Trainera.