2 December 2026: the AI Act deadline your website must meet

From 2 December 2026, AI-generated content from systems that were already on the market must be machine-readably marked under Article 50(2) of the EU AI Act. The legal duty falls on providers of generative AI systems, but if your website publishes AI-generated text, images or video, the change still reaches you. Your tools have to mark their output, and your site has to keep that marking intact. This article explains what changes, who has to act, and gives you a 10-point checklist to work through before the date.

What changes on 2 December

The AI Act already requires transparency since 2 August 2026: you must tell visitors when they talk to a chatbot, and you must label deepfakes. From 2 December 2026, the next piece applies. Generative AI output has to carry a machine-readable mark, so that other systems can recognise it as AI-generated. The rule covers content from systems placed on the market before 2 August 2026. Providers are already preparing: xAI, for example, said Grok API text will carry an invisible watermark from 7 November 2026. Expect your own tools to follow the same pattern, with settings or watermarks you may need to switch on.

Who has to act

If you run a website or an online shop in Germany, check three places. First, your content: blog posts, product descriptions, images or videos made with AI. Second, your chatbot: if it answers customers, it must say so, and it must not pretend to be a person. Third, your tools: the AI services you use are the providers, and they carry the marking duty, but you are the one who publishes the result. If a tool strips the marking when you download or re-upload a file, the chain breaks on your site. In Germany, the Bundesnetzagentur has supervised AI under the AI Act since 29 July 2026.

A 10-point checklist for your website

1. List every place AI touches your site: text, images, video, chatbot, product descriptions, translations.
2. For each tool, check whether it marks AI output in a machine-readable way, and switch the setting on where it exists.
3. Add a short, visible note where AI content appears, for example "Created with AI, reviewed by our team."
4. Make sure your chatbot introduces itself as AI in its first message.
5. Label any deepfake or synthetic video clearly, and keep a record of what you published.
6. Check your publishing chain: does the marking survive when you download, resize or re-upload a file? Test one file end to end.
7. Tell your team who is allowed to publish AI content, and what the note has to say.
8. Keep a simple log: which tool made which piece of content, and when.
9. Review your privacy notice and terms where they mention automated content.
10. Put a reminder in your calendar for the next dates below, so this is not a one-off fix.

What comes after December

The 2 December date is not the end of the story. The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026, moved the high-risk duties to 2 December 2027 for systems in Annex III and 2 August 2028 for AI in regulated products, and it eased the rules for small and medium-sized businesses. Two more dates matter for German businesses: from 9 December 2026, software and AI count as products under the new Product Liability Directive, so missing security updates can make a product defective. And from 1 January 2027, German businesses with turnover above 800,000 euro must issue e-invoices.

What to do this week

Start with the inventory in step 1 of the checklist. Most small businesses find the list is shorter than they feared, and the fixes are small: a setting here, a short note there, a test upload. The expensive mistake is to ignore the date and discover later that your chatbot has been answering customers without saying it is AI. Do the inventory this week, switch on the marking settings, and add the visible notes. If you want help, our AI content engine can set up a workflow where AI drafts and your team reviews, and our AI chatbot service builds chatbots that introduce themselves properly. For the bigger picture, read our guide on Germany's KI-Nation 2030 plan.

Takeaway

2 December 2026 is a marking deadline, not a ban. Label your AI content, check that the marking survives publishing, and tell your visitors honestly when they are reading or talking to AI. That is the whole job, and it is doable in a week.