/

Updated 18 November 2026

AI Transparency Rules Every Website Must Follow (EU AI Act Article 50)

AI Disclosure: This content includes AI-generated text.

Short answer: the EU AI Act’s transparency rules for websites are in Article 50, and they have applied since 2 August 2026. For a typical business website they come down to four duties: tell people when they’re talking to an AI, label deepfakes, label AI-written public-interest text that nobody reviewed, and present those notices clearly and accessibly. Separate EU laws add a few more transparency duties for personalised prices, recommendations and automated decisions.

This guide lists every rule that commonly applies to a website, who carries it, and what it means in practice. It’s based on the text of the AI Act as amended by the Digital Omnibus (Regulation (EU) 2026/1744) and the Commission’s Article 50 guidance.

The rules at a glance

RuleApplies toWebsite exampleSince
Art. 50(1) AI interaction noticeProvider of the AI systemChatbot, voice assistant2 Aug 2026
Art. 50(2) machine-readable markingProvider of generative AIWatermarks in images from AI tools2 Aug 2026 (2 Dec 2026 for systems already on the market)
Art. 50(3) emotion recognition and biometric categorisation noticeDeployerRare on websites: webcam mood analysis, face-based age estimation2 Aug 2026
Art. 50(4) deepfake disclosureDeployerRealistic AI photo of a person, place or event2 Aug 2026
Art. 50(4) public-interest text disclosureDeployerUnreviewed AI article on a law, health or finance topic2 Aug 2026
Art. 50(5) how to discloseEveryone aboveClear, at first exposure, accessible2 Aug 2026
Art. 4 AI literacyProviders and deployersStaff who use AI tools know their limits2 Feb 2025

A provider builds an AI system or puts one into service under its own name. A deployer uses one professionally. A business website using ChatGPT, Midjourney or a chatbot widget is usually a deployer, but if you build your own assistant on a model API and run it under your brand, you can become its provider. Our guide on responsibility for third-party AI tools covers where the line falls.

Rule 1: say when people are talking to an AI

Anyone who chats or talks with an AI system must be told it’s an AI, unless that’s obvious to a reasonably well-informed, observant person. In practice: the chatbot’s first message and header say “AI assistant”, a persona name is fine only if it’s paired with that, and switches between human and AI agents are announced. Full wording examples are in our chatbot compliance guide.

Rule 2: AI tools must mark their output (the provider’s job)

Providers of systems that generate images, audio, video or text must mark the output in a machine-readable way, for example with embedded metadata or watermarks, so it can be detected as AI-generated. The Digital Omnibus gave providers whose systems were already on the market before 2 August 2026 until 2 December 2026 to comply. This is not a website owner’s duty, and it doesn’t replace your own visible disclosure. Don’t strip that metadata from images on purpose, though: it’s part of how the system is meant to work.

Rule 3: tell people about emotion recognition or biometric categorisation

If you use a system that infers emotions or sorts people into categories from biometric data, the people exposed must be informed, and GDPR rules on biometric data apply in full. Most websites don’t do this. Watch for it in video-interview tools, webcam “try-on” features or age-estimation checks.

Rule 4: label deepfakes

If you publish AI-generated or manipulated images, audio or video that resemble real people, objects, places, entities or events and would falsely appear authentic, you must disclose it. Illustrations and obvious fantasy don’t count. Photoreal “team photos” of people who don’t exist, or a product photo showing features the product doesn’t have, do. For artistic, satirical or fictional work, a lighter disclosure that doesn’t spoil the work is enough.

Rule 5: label unreviewed AI text on public-interest topics

AI-generated text published to inform the public on matters of public interest must be disclosed, unless it went through human review or editorial control and a person or company holds editorial responsibility. Public interest covers topics like politics, law, public health, finance, the environment and consumer safety. Product descriptions and ordinary marketing copy usually fall outside it. Our Article 50 test for AI content walks through the decision.

Rule 6: make every notice clear, timely and accessible

Article 50(5) sets the form for all of the above: clear and distinguishable, given at the latest at first interaction or exposure, and compliant with accessibility requirements. That rules out notices hidden in terms, footers, hover tooltips or image pixels with no text alternative. See where to place AI disclosures for patterns by content type.

Rule 7: make sure your team understands the AI they use

Article 4 has applied since 2 February 2025. After the Digital Omnibus, it requires providers and deployers to take measures to support the AI literacy of staff and others operating AI on their behalf, without demanding a specific level. For a small site, that means the people who write prompts, publish AI content or configure the chatbot know what these tools get wrong and how to check them. A short internal guide and a review habit are a reasonable start.

Transparency rules from other EU laws

Website AI often triggers duties outside the AI Act. The common ones:

  • Personalised prices. Under EU consumer law, if a price was personalised on the basis of automated decision-making, you must tell the consumer before the purchase.
  • Recommendations and rankings. Online marketplaces must explain the main parameters that determine product rankings, and online platforms under the Digital Services Act must explain the main parameters of their recommender systems.
  • Automated decisions about people. GDPR requires you to inform people about automated decision-making with legal or similarly significant effects, explain the logic involved, and, in many cases, offer human review.
  • Personal data in AI tools. Your privacy notice should name AI tools that process visitor data, such as chatbots and personalisation engines. Our EU AI Act vs GDPR comparison explains how the two laws overlap.

What doesn’t need a label

  • Text where AI only fixed grammar, spelling or style.
  • AI-written content on topics that aren’t of public interest, such as most product and marketing copy.
  • Reviewed AI text on public-interest topics, where a named person or company takes editorial responsibility. Keep a record of the review.
  • Stylised or obviously artificial images that nobody would mistake for a real photo.
  • AI used behind the scenes, such as spam filtering or search ranking, where nobody interacts with it or sees generated content. Other laws may still apply.

Penalties

Breaches of Article 50 can be fined up to 15 million euros or 3% of worldwide annual turnover. For SMEs and start-ups the lower of the two caps applies. National market surveillance authorities enforce it, and fines must be proportionate. Details in EU AI Act fines and penalties explained.

Putting it into practice

  1. List every AI feature and every type of AI content on your site. Our website AI audit has a template.
  2. Map each one to the rules above. Most sites end up with two or three that actually apply.
  3. Add the notices and test them on mobile and with a screen reader.
  4. Write down your decisions, including where you concluded no label was needed and why.
  5. Review every few months, and whenever you add a new AI tool.

On WordPress, the free EU AI Act Ready plugin handles the visitor-facing part: per-post disclosure levels, notices only where needed, AI image labels, chatbot notices for common widgets, and an exportable record of your decisions. It runs on your own server and needs no account.

Frequently asked questions

Do all websites using AI need an AI disclosure?

No. Only the situations above trigger a legal duty. A site that uses AI to draft product descriptions and reviews them has, in most cases, nothing to label under Article 50.

Did the Digital Omnibus delay the transparency rules?

Mostly no. It delayed high-risk AI obligations to December 2027 and August 2028. For Article 50, it only gave providers of generative systems already on the market until 2 December 2026 for machine-readable marking. Chatbot, deepfake and text disclosure duties have applied since 2 August 2026.

Is there an official AI label I can use?

The Commission has published a common set of EU icons for labelling AI-generated content, and the Code of Practice on transparency of AI-generated content recommends them. Using them is optional. Pair the icon with text.

Do these rules apply to websites outside the EU?

Yes, if the AI system’s output is used in the EU. A site serving EU visitors should assume Article 50 applies regardless of where the business is based.

Practical guide: Article 50 explained for websites, paragraph by paragraph.