theuxbites
EU AI Act · Article 50

Which AI transparency rules apply to you?

Article 50 of the EU AI Act has applied since 2 August 2026. It sets four transparency obligations, the last of which has two parts, so this check looks at five things: telling people they are talking to AI, marking AI-generated output and making it detectable, informing people exposed to emotion recognition, labelling deep fakes, and disclosing AI-written text on public-interest matters. Answer up to twenty questions and see which of them your answers trigger, with the paragraph each one rests on.

What this check is, and is not

  • It applies fixed rules to your answers. No AI model is used, and your answers never leave your browser.
  • It tells you which obligations apply. It cannot tell you whether you comply — it has not seen your product.
  • It covers Article 50 only. It does not assess prohibited practices, high-risk systems, general-purpose model duties, law-enforcement exceptions, the GDPR, or the AI-labelling rules of social platforms.

Guidance, not legal advice

This check is support, not a ruling. It works only from the official EU sources listed with it, as they read on the date shown, and it is not kept up to date. It helps you recognise what may need a closer look. It is not legal advice, it does not establish whether you meet the AI Act, and theuxbites takes no responsibility for decisions based on it. Take the result to a lawyer or another qualified professional before you act on it.

Independent, and not the official tool

theuxbites is an independent product. This check is not an official EU tool and is not affiliated with, endorsed by or reviewed by the European Commission or any authority. The official tool is the European Commission’s official AI Act Compliance Checker, which looks at the AI Act as a whole. This check looks only at Article 50, one question at a time. They are meant to be used together: if the two ever point in different directions, go with the official sources and ask a lawyer.

Scope

Whether the AI Act reaches you at all.

Are you established in the European Union, or is your product or the content you publish used or seen by people there?

Either is enough. The AI Act covers providers and deployers established in the EU, and those outside it when the output of the AI system is used in the EU — for example a product offered to EU customers, or content posted on the open internet. Use in the EU that is incidental and that you could not foresee is not meant to trigger it on its own.

AI in your product

AI features your own users interact with or receive output from.

Does your product have an AI feature that your users interact with, or that produces content for them?

For example a chatbot, an assistant, generated text, generated images, AI summaries or AI recommendations written in natural language. Using AI only internally to write your own code or emails does not count here. A purely rule-based bot with scripted answers is not an AI system.

AI in what you publish

Images, audio, video and text you put in front of the public — website, ads, social media, newsletters.

Do you publish images, audio or video that were generated or substantially altered with AI?

On your website, in ads, on social media or in your product — made by you, or by others under your direction and control. Routine adjustments to real footage — cropping, colour correction, lighting, noise removal, removing a passer-by in the background — are not a substantial alteration.

Do you publish text to the public that was generated or substantially rewritten with AI?

Blog posts, articles, reports, newsletters, social posts — anything a large, open-ended audience can read, paid or free. Spell-checking and light editing of your own writing do not count. Private correspondence, a small closed group, and internal documents are not published.

Emotion recognition and biometrics

Rare, but it carries its own obligation.

Do you use AI to recognise people’s emotions, or to sort people into categories based on biometric data?

For example inferring mood or intentions from a face or voice, or assigning people to categories such as age group or gender from their biometric data. A feature that is purely ancillary to another service and strictly necessary for technical reasons is not biometric categorisation.

0 of 5 questions answered.

Read the rules yourself

This check is meant to help you find your way, not to stand between you and the source. These are the official EU pages. If a question goes beyond what a questionnaire can answer, the Commission’s service desk and your own lawyer are the right places to take it.

Need something more specific?

Write to David Rozsa, who built theuxbites, at hello@theuxbites.com. He will tell you plainly whether it is something he can help with, or whether it belongs with a lawyer who practises in this area.

What this check is based on

The Regulation is the law. The Commission guidelines explain how the Commission reads it and do not bind courts or authorities. The Code of Practice is voluntary; following it is one recognised way to show the marking obligation is met, not the only one.

This check reflects the sources as read on 3 October 2026. It is not updated automatically and may be out of date when you read it — the official pages above are always the current version. theuxbites publishes its own Article 50 statement at AI Transparency.