# AI Act Article 50: What Website Owners Must Do From 2 August

AI Act Article 50 applies from 2 August 2026. A practical checklist for chatbot widgets, AI-written articles and AI images on your website.

By Jasper Koers · Published 2026-08-03

## Key takeaways

- The transparency obligations in Article 50 of the EU AI Act apply from 2 August 2026 to providers and deployers of chatbots, generative AI systems, emotion recognition systems and deepfakes.
- Under AI Act Article 50(1), a chatbot or AI agent that talks directly with people must make clear from the first interaction that the user is dealing with AI, unless that is obvious.
- Under AI Act Article 50(4), AI-generated text published to inform the public on matters of public interest must be labelled, unless it has undergone human review or editorial control and someone holds editorial responsibility for it.
- According to the European Commission, only the machine-readable marking duty in AI Act Article 50(2), and only for AI systems already on the market before 2 August 2026, gets extra time until 2 December 2026.
- Breaches of the AI Act Article 50 transparency obligations can be fined up to 15 million euros or 3% of worldwide annual turnover under Article 99(4) of the AI Act, with lower caps for SMEs.

Since yesterday, **2 August 2026**, the transparency obligations in **Article 50 of the EU AI Act** ([Regulation (EU) 2024/1689](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)) apply. If your website has a chatbot widget, publishes AI-written articles or uses AI-generated images, some of these rules now apply to you: visitors must be told when they talk to an AI, deepfakes must be disclosed, and some AI-generated text must carry a label.

The European Commission published its [guidelines on the Article 50 transparency obligations](https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems) on 20 July 2026 (Communication C(2026) 5054), plus an [FAQ](https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act) that answers most practical questions. This guide turns them into a checklist for website owners.

> **Not legal advice.** This article summarises the AI Act and the Commission's guidance as of 3 August 2026. Check your own situation with a lawyer or your compliance team.

## What AI Act Article 50 covers

Article 50 is the AI Act's "transparency risk" layer. It does not ban anything and it does not depend on whether your AI system is high-risk. It simply says that people should know when AI is involved. It has five parts:

- **Article 50(1):** AI systems that interact directly with people, such as chatbots, AI agents and avatars, must be designed so people know they are dealing with AI.
- **Article 50(2):** generative AI systems must mark their output (text, image, audio, video) in a machine-readable way, so it can be detected as AI-generated or manipulated.
- **Article 50(3):** deployers of emotion recognition or biometric categorisation systems must inform the people exposed to them.
- **Article 50(4):** deployers must disclose deepfakes, and must label AI-generated text published to inform the public on matters of public interest.
- **Article 50(5):** the information must be clear and distinguishable, and given at the latest at the first interaction or exposure.

## Provider or deployer: which role do you have?

The AI Act splits the duties between two roles, and the split matters for a website owner.

- A **provider** develops an AI system (or has one developed) and places it on the EU market or puts it into service under its own name. The company that sells you a chatbot product, or builds an image generator, is a provider.
- A **deployer** uses an AI system under its own authority in a professional context. If you put a vendor's chatbot on your shop or publish AI-generated images on your company blog, you are a deployer. According to the Commission's FAQ, employees acting for a company are not separate deployers; the company is.

If you build your own chatbot on top of a model API and run it under your brand, you can be **both**: the provider of that chatbot system and its deployer.

## The AI Act transparency obligations as a website checklist

| Obligation | Who carries it | What a website owner should do |
| --- | --- | --- |
| Chatbot disclosure, Art. 50(1) | Provider of the chatbot system | Check that your chatbot widget tells visitors it is AI at the start of the conversation. If you built it yourself, this duty is yours. |
| Machine-readable marking, Art. 50(2) | Provider of the generative system | Mostly your tool vendor's job. Don't strip the metadata or watermarks from AI images and videos you publish. |
| Emotion recognition / biometric categorisation, Art. 50(3) | Deployer | Rare on websites. If you use it (for example in a video call tool), inform the people exposed. |
| Deepfake disclosure, Art. 50(4) | Deployer | Disclose AI-generated or manipulated images, audio or video that look real and could pass for authentic. |
| Labelling AI text on public-interest matters, Art. 50(4) | Deployer | Label AI-written text published to inform the public on matters of public interest, unless a person reviewed it and someone holds editorial responsibility. |

### AI chatbot disclosure: the most common case

A support or sales chatbot is the Article 50 issue most websites will actually have. The Commission's FAQ says the duty applies when four criteria are all met: the system is AI, there is a genuine two-way exchange, the AI communicates directly (not through a human), and the other side is a natural person.

The disclosure must come **from the start of the first interaction**, unless it is obvious from the context that you are talking to an AI. The Commission says that exception is to be read restrictively, so don't rely on it because your bot has a robot icon. In practice:

- Put a short line in the chat window before the first message, such as "You're chatting with an AI assistant."
- Make it readable for everyone, including screen-reader users. Article 50(5) ties the information to the EU accessibility requirements.
- If a human can take over the conversation, say when that happens.
- Check your vendor's settings. Most chatbot products are provided by a vendor, and Article 50(1) puts the design duty on the provider, but you are the one who configures the widget and you should not switch the notice off.

### How to label AI-generated content on your site

Two separate rules apply to deployers under Article 50(4).

**Deepfakes.** If you publish an AI-generated or manipulated image, audio or video that resembles real people, objects, places or events and would falsely appear authentic, you must disclose that it is AI-generated, "upon first exposure at the latest" according to the FAQ. A provider's invisible watermark does not satisfy this duty; the disclosure must be perceivable. For evidently artistic, satirical or fictional work, the disclosure may be done in a way that does not spoil the work.

A stylised illustration for a blog header is usually not a deepfake: it does not create a false appearance of authenticity. A realistic "photo" of a product being used by customers who don't exist is a much closer call.

**AI-generated text.** Text generated or manipulated by AI and published "with the purpose of informing the public on matters of public interest" must be labelled. The Commission gives politics, public health and consumer safety as examples of qualifying topics. The label is **not** needed if the text has undergone human review or editorial control, and a person or organisation holds editorial responsibility for publication. The FAQ is strict about what counts: a deliberate review of the substance by a qualified person, not a spell-check.

For most company blogs, that means: if a person edits and approves AI-assisted articles before they go live, and your company stands behind them, Article 50(4) does not require a label. If you auto-publish AI-written news or health content without review, it does.

### What your vendors must do

Article 50(2) is aimed at the providers of generative AI systems: they must mark output in a machine-readable format that is, in the Commission's words, "effective, reliable, robust and interoperable". The Commission also published a voluntary **Code of Practice on Transparency of AI-Generated Content** that providers and deployers can sign to show compliance with Article 50(2), (4) and (5).

As a website owner you don't have to build watermarking, but it is worth asking your chatbot and image-tool vendors how they comply, and making sure your publishing pipeline (image compression, CMS uploads) does not remove those markings where it can be avoided.

## Dates, grace periods and enforcement

| Date or topic | What applies |
| --- | --- |
| 2 August 2026 | Article 50 applies, for all four paragraphs with duties. |
| 2 December 2026 | Deadline for the Art. 50(2) marking duty, **only** for generative AI systems placed on the market before 2 August 2026. |
| Content made before 2 August 2026 | No retroactive labelling required, although the Commission encourages it. |
| Enforcement | National market surveillance authorities, with a limited role for the EU AI Office. |
| Fines | Up to €15 million or 3% of worldwide annual turnover, whichever is higher ([Article 99(4)(g)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)); for SMEs and start-ups, whichever is lower. |

Note what the grace period does **not** cover: chatbot disclosure, deepfake disclosure and text labelling apply now, also for systems that were already running before 2 August.

## Your AI Act website checklist

1. **List your AI features.** Chatbots, AI search on your site, AI-generated images, AI-written articles, voice or avatar tools.
2. **Decide your role for each.** Vendor product (you are the deployer) or built in-house (you are provider and deployer).
3. **Add a chatbot notice** at the start of every conversation, visible and accessible.
4. **Write down your editorial process** for AI-assisted content: who reviews, who approves, who is responsible.
5. **Label realistic AI images and video** that could pass for real.
6. **Ask vendors** how they meet Article 50(2) and whether they signed the Code of Practice.
7. **Mention AI use in your privacy notice** if your chatbot processes personal data. That is a GDPR duty, not an AI Act one, and it applies alongside Article 50.

## Measuring your chatbot without adding a privacy problem

Many teams want to know whether their chatbot is used and whether it helps conversions. That is reasonable, but avoid solving an AI Act task by creating a GDPR one: counting chat opens or "chat started" events does not require logging who the visitor is, storing chat transcripts in your analytics tool or setting tracking cookies. Our [GDPR guide for web analytics](/blog/gdpr-compliance-for-web-analytics-the-complete-guide) covers the basics, and if you run a SaaS product, the [privacy-compliant SaaS analytics setup](/blog/setting-up-privacy-compliant-analytics-for-your-saas) shows how to track product events without personal data.

## What this means for your analytics

Article 50 is about telling people when AI is involved, not about analytics. Still, the principle is the same one we built ClearAnalytics on: be clear about what happens on your site and collect no more than you need. ClearAnalytics sets no cookies, stores nothing on the visitor's device, processes data in the EU and can count custom events such as chatbot opens without identifying anyone. See [how it works](/features).

## Frequently asked questions

### Does my website chatbot need to say it is AI under the AI Act?

Yes, in most cases. From 2 August 2026, Article 50(1) requires that people are told they are interacting with an AI system at the latest at the first interaction, unless that is obvious to a reasonably well-informed person. The Commission says that exception is read restrictively.

### Do I have to label AI-written blog posts under the AI Act?

Only AI-generated text published to inform the public on matters of public interest must be labelled, and not if a person reviewed it and someone holds editorial responsibility. A product page or a how-to article generally falls outside that rule, but a misleading undisclosed deepfake image never does.

### When does AI Act Article 50 apply?

From 2 August 2026. Providers of generative AI systems already on the market before that date get until 2 December 2026 for the machine-readable marking duty in Article 50(2) only.

### Who enforces AI Act Article 50?

Mainly the national market surveillance authorities in each EU member state. The EU AI Office has a limited role for systems built on a provider's own general-purpose model and for systems inside very large online platforms or search engines.

### What are the fines for breaching the AI Act transparency obligations?

Up to 15 million euros or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher, under Article 99(4)(g). For SMEs and start-ups the lower of the two amounts applies.

Source: https://clearanalytics.eu/blog/ai-act-article-50-website-chatbots
