Labelling requirements under the European AI Regulation (EU AI Act)

The situation: AI is everywhere – but it is almost never labelled as such

In theory, the situation is clear: anyone using artificial intelligence knows that they are using artificial intelligence. In practice, however, this is often not apparent to website users. A chatbot responds as fluently as a human, a product image looks like a real photograph, a telephone helpline sounds like a real customer service representative – and nobody realises that an AI system is at work in the background.

This is precisely where the European AI Regulation (EU AI Act) comes in. From 2 August 2026, the transparency obligations set out in Article 50 will become binding. Essentially, they require something simple yet far-reaching: people must be able to recognise when they are dealing with AI. What was long considered a matter of good practice has thus become a matter of legal compliance.

For you, as a website operator, this means that the use of chatbots, AI-generated images, voicebots or AI-generated text remains possible and sensible – it is simply no longer without conditions. This article sets out what matters now.

Why transparency is no longer an optional feature

The rationale behind Article 50 is not a technical one, but a social one: trust is built when people know what they are dealing with. Content that looks deceptively real and bots that appear human pose a risk of manipulation – ranging from harmless misunderstandings to deliberate misrepresentation. The Regulation addresses this risk by requiring disclosure.

It is important to understand that these obligations apply not only to providers of AI systems, but also explicitly to operators – that is, those who use an AI system in their own name. And that is usually you, as soon as you integrate a chatbot into your website or publish AI-generated content. Responsibility cannot be delegated entirely to the tool provider.

In short: labelling is no longer a matter of choice, but of obligation.

An overview of the three mandatory areas

Article 50 divides the transparency obligations into several areas. Three of these are particularly relevant for website use: artificially generated or manipulated media, interactive systems such as chatbots and voicebots, and AI-generated text.

Images, videos and audio: The issue of deepfakes

Any image, audio or video material that has been generated or altered using AI and appears to be a deepfake or a deceptively realistic simulation of real people, places or events must be clearly labelled as artificially generated or manipulated . Therefore, as soon as there is a risk that content might be perceived as depicting a real scene, the labelling requirement applies.

However, not every AI-generated image is affected. Exceptions include, in particular:

• purely decorative elements or simple graphics that pose no risk of misleading the viewer
• purely private uses with no public relevance
• art, fiction and satire – in these cases, the disclosure may be made discreetly so as not to detract from the work, but it must be present

A visible label alone often does not meet the requirements of the regulation. A machine-readable label – via watermarks or metadata – is also required, so that platforms, search engines and other systems can automatically recognise AI-generated content. This reduces the scope for misuse and ensures traceability beyond the individual channel. You should therefore check whether the tools you use already automatically apply such watermarks.

Chatbots and voicebots: clarity before the first word

This point directly affects most website operators. If you use an AI-powered chatbot or voicebot , users must be informed at the start of the conversation that they are communicating with an AI rather than a human. This notification must be clear, unambiguous and provided before the actual interaction begins – not hidden away in the terms and conditions, and not only after several messages have been exchanged. In practice, a straightforward introduction is usually sufficient:

• Chatbot: “Hello! I am [your company’s] virtual assistant and I answer your questions automatically. If you wish, I can put you through to a member of our team.”
• Voicebot: “You are speaking to an automated voice assistant.” An exception applies where it is already obvious to a reasonably attentive person that an AI is involved. However, when in doubt, a clear disclosure is always the safest option.

Consequences of non-compliance

The disclosure requirements are not merely a non-binding request. Failure to comply with the transparency requirements may result in fines of up to 15 million euros or up to 3 per cent of global annual turnover – whichever is higher. For most small and medium-sized enterprises, it is not so much the absolute upper limit that is decisive as the clear message behind it: the EU is serious about transparency, and this issue should not be put on the back burner.

The path to a compliant website: from the current state to the target state

In most cases, the effort involved in ensuring proper labelling is manageable – provided you take a structured approach. We recommend following a clear, step-by-step process:

• Assessment: Systematically identify where AI is used on your website – chatbots, image generation, product descriptions, voicebots, translations.
• Check bots: Ensure that chatbots and voicebots clearly indicate their AI nature right from the start.
• Check media: Label AI-generated or AI-edited images and videos that appear realistic.
• Check technology: Clarify whether your tools add watermarks and metadata, and enable these features where available.
• Define responsibilities: Appoint someone to monitor this area when new AI features are introduced.
• Document: Keep a clear record of the measures implemented.

Transparency builds trust

An often-overlooked point: labelling is not just an obligation, but also an opportunity. Communicating openly that a chatbot is a chatbot and that an image has been generated by AI signals honesty. Particularly at a time when users are becoming increasingly sensitive to AI, transparency becomes a factor in building trust. A compliant website is therefore not the toothless option, but the more credible one.

Useful resources

• The full text of Article 50 of the AI Regulation: transparency obligations under Article 50 and the official version of the EU AI Act
• EU icons for labelling: EU icons for AI-generated content
• European Commission Code of Conduct with practical guidance on implementation

Conclusion: Labelling as a ‘trust case’ – not just a formality

The transparency obligations set out in Article 50 are more than just a regulatory footnote – they mark a shift in attitude. Instead of using AI behind the scenes, you will make its use visible in future. This is, first and foremost, an obligation that can be met with manageable technical effort: a clear notice on the bot, a label on AI-generated content, activated watermarks, and documented responsibilities.

Those who take this step proactively will significantly reduce their legal risk – whilst simultaneously gaining credibility with their own customers. Transparency costs little and pays off twice over: in terms of legal certainty and trust. This is precisely what makes labelling a clear business case and not just a compliance tick-box exercise.

If you’d like to implement labelling on your website in a technically sound manner, we’d be happy to support you – from the initial assessment right through to integration.