Labelling AI content: chatbots, images and texts

What Article 50 requires and since when

Article 50 of the AI Act, Regulation (EU) 2024/1689, applies from 2 August 2026 (Article 113). Regulation (EU) 2026/1744 kept this date and added one transition rule: providers of AI systems that generate synthetic content and were placed on the market before 2 August 2026 have until 2 December 2026 to mark the output in a machine-readable way (Article 111(4)).

Article 50 distinguishes two roles. A provider develops an AI system, or has it developed, and places it on the market under its own name. A deployer uses an AI system under its authority, except for personal non-professional use. If you add a chatbot service or an image generator to your website, the company behind the service is usually the provider and you are the deployer. If you build the chatbot yourself or offer it under your own name, you can be the provider.

Chatbots: say that an AI answers

Article 50(1): AI systems that interact directly with people must be designed so that people are informed that they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person. This is the duty of the provider. As the operator of the website you choose the service and its settings, so check whether the chat says clearly that an AI answers. A human first name and a photo make it less obvious that a machine is talking.

Images, audio and video: deep fakes

A deep fake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (Article 3(60)). Deployers must disclose that such content was artificially generated or manipulated. If it is part of an evidently artistic, creative, satirical or fictional work, it is enough to disclose that such content exists in a way that does not hamper the display or enjoyment of the work (Article 50(4)).

A realistic AI photo of a team that does not exist or of a product in a real place can fit this definition. An obviously drawn illustration usually does not look authentic. Decide case by case, and if in doubt, add a label.

Machine-readable marking

Under Article 50(2), providers of systems that generate synthetic audio, images, video or text mark the output in a machine-readable format, except for tools that only assist with standard editing. What you can do: keep this metadata. Google describes the IPTC field DigitalSourceType with the value trainedAlgorithmicMedia for AI images and can show C2PA information in the ‘About this image’ feature. For Merchant Center, Google requires this IPTC value for AI-generated product images. Check whether your image optimiser keeps the metadata.

AI texts on matters of public interest

Deployers who publish an AI-generated or manipulated text to inform the public on matters of public interest must disclose this. The duty does not apply if the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication (Article 50(4), second subparagraph). Article 50 does not define matters of public interest. Whether a blog post or guide falls under it depends on its content.

When and how to show the notice

The information must be given in a clear and distinguishable manner at the latest at the time of the first interaction or exposure, and it must meet the applicable accessibility requirements (Article 50(5)). In practice:

• Chat: in the first message, before the visitor types anything.
• Image: right at the image, as a visible label or caption, and in the alt text where it helps.
• Video and audio: at the start, and again in the description.
• Text: below the headline, not only at the end of the page, as real text with enough contrast.

Templates you can adapt

Replace the parts in square brackets. The templates are examples. They do not decide whether a duty applies to you. Chatbot, first message:

Hello, I am [name], the AI assistant of [company].
I am an AI system, not a person, and I can make mistakes.
To talk to a person, write "team" or call [phone number].

Label for a realistic AI image or video:

AI-generated image
This video was created with AI. The people shown are not real.

Notice for an artistic, satirical or fictional work, for example in the credits:

Some scenes in this video were created with AI.

Notes for AI texts, reviewed and not reviewed:

This article was drafted with the help of AI. [Name] from our editorial team checked and edited it and is responsible for its content.
This text was generated with AI and has not been reviewed by a person.

Fines and national rules

Member States lay down the rules on penalties and take into account the interests of small and medium-sized enterprises (Article 99(1)). Breaches of the transparency obligations in Article 50 can lead to administrative fines of up to 15 million euros or, for companies, up to 3 % of worldwide annual turnover, whichever is higher (Article 99(4)). Which authority is responsible in Germany or Poland is not covered in this guide.

Work with Do4U

The AI transparency check reads a page address or pasted HTML and asks three short questions: do you use a chatbot, do you publish AI images or videos, and do you publish AI texts on topics of public interest? The code looks for chat widgets of known services, signs of AI-generated media in the HTML and existing AI notices. The AI sorts out which situations likely apply, with the paragraph and a reason, and writes short notice texts. The code builds the HTML for a visible label. It cannot see a chat that a script loads later and never opens image or video files.

Do4U finds hints and builds drafts. It never says that your website meets the AI Act. You check each point and decide.