EU AI Act 2026: What Website Chatbots Must Make Transparent from August 2

As of August 2, 2026, a good UX practice becomes a concrete transparency obligation: People must be able to recognize when they are interacting with an AI system. For website chatbots, now is therefore the right time to check labeling, greetings, accessibility, and internal evidence.

The deadline is particularly current: On July 9, 2026, the EU Commission published its assessment of the new transparency code for AI-generated content. At the same time, Article 50 of the EU AI Act is moving into the operational daily routine of companies. This article explains what website operators should practically check regarding an AI chatbot. It is an editorial orientation and not legal advice.

The short answer: What changes on August 2, 2026?

The EU AI Regulation generally applies from August 2, 2026. Article 50 contains special transparency obligations for certain AI systems. For chatbots, paragraph 1 is particularly relevant: providers must design interactive AI systems so that affected persons are informed that they are interacting with an AI system. An exception exists only if this is obvious from the perspective of a reasonably informed, attentive, and discerning person.

Additionally, Article 50 paragraph 5 requires that this information be provided clearly and distinguishably, at the latest during the first interaction or exposure, and in compliance with applicable accessibility requirements.

In practice, this means: A small robot avatar alone is not a reliable compliance strategy. The labeling should be linguistically unambiguous and become visible or perceptible before users can mistake the chat for a human service channel.

Why this topic is not simply the GDPR article in new packaging

Data protection and AI transparency overlap but pursue different goals. The GDPR concerns, among other things, personal data, legal bases, information obligations, storage duration, and data subject rights. Article 50 of the AI Act, on the other hand, addresses transparency regarding the artificial nature of the system and certain content.

A data-protection-compliant chatbot is therefore not automatically AI Act-compliant. Conversely, a visible notice saying "AI Assistant" replaces neither a privacy policy nor data minimization. Website operators should document both audit paths separately and then merge them into a comprehensible user guidance process.

Who is responsible: provider, operator, or both?

The AI Act distinguishes between different roles. Simplified, the provider is the organization that develops an AI system or brings it to market under its own name. The operator uses the system under its own responsibility. In the case of a SaaS chatbot, software providers and website companies may therefore bear different obligations.

The obligation from Article 50 paragraph 1 is explicitly addressed to providers: the system must be technically designed so that the information is possible and effective. Website operators should nevertheless check whether the labeling remains actually visible in their specific integration. A correctly prepared widget helps little if theme CSS, a renamed header, or a customized greeting covers the notice again.

Which role a company assumes in an individual case depends on the product, branding, customizations, and contractual design. In case of uncertainty, responsibilities should be clarified with the provider and, if necessary, legally.

Practical checklist for website chatbots

1. Clearly label the chatbot as AI

Use a clear designation such as "AI Assistant" or "AI Chatbot". Terms such as "Support", "Consultant", or a human first name can create the impression that a human is answering without additional notice.

Practical example for the widget header: "AI Assistant from Sample Company"

2. Provide the notice at the latest during the first interaction

The information should not appear only on a subpage, in a tooltip, or at the end of a long greeting. A combination of visible designation in the header and a short sentence in the first message is sensible.

Practical example for the greeting: "Hello, I am the AI Assistant from Sample Company. I answer questions based on our approved website content."

3. Do not solve accessibility through color alone

A colored dot or a symbol without text is not sufficient for many users. The notice should be available as actual text for screen readers, have sufficient contrast, and be reachable in a sensible order during keyboard operation. Check desktop, smartphone, zoom, and common screen reader scenarios.

4. Name the limits of the system understandably

Transparency means more than the word "AI". Users should be able to recognize what answers are based on and when a human review is sensible. Especially regarding prices, contracts, health, safety, or individual legal questions, the chatbot should not feign certainty that it cannot deliver.

5. Offer a traceable path to a human

The AI Act does not generally prescribe a live chat for every website chatbot. Nevertheless, an escalation option is a strong trust and risk measure. Contact forms, email, callback options, or live chat should be offered where the bot cannot answer a question reliably.

6. Keep knowledge sources and approvals up to date

A clearly labeled chatbot can still give wrong or outdated answers. Therefore, maintain FAQs, documents, and website content, check recurring wrong answers, and define who approves changes. Transparency and answer quality belong together in practice.

7. Document the implementation

At a minimum, record when the labeling was introduced, which texts users see, which languages are covered, how accessibility was checked, and who is responsible for regular checks. Screenshots and version states help make later changes traceable.

What the new EU transparency code covers – and what it does not

The code published on June 10, 2026, primarily concerns the machine-readable marking of AI-generated content as well as the labeling of deepfakes and certain text publications. In July 2026, the EU Commission and the AI Board classified it as a suitable voluntary instrument to support the obligations from Article 50 paragraphs 2, 4, and 5.

Important for website chatbots: The direct information that a user is interacting with AI is found in Article 50 paragraph 1. The code for AI-generated content does not replace this notice. Companies should therefore answer two questions separately:

• Is the interaction with the AI chatbot clearly recognizable for users?
• Must generated or published content be additionally marked or disclosed?

For AI-generated texts on matters of public interest, Article 50 contains an important exception if a human review or editorial process has taken place and a natural or legal person bears the editorial responsibility. Especially in automated blog processes, this review step should not only be planned but actually carried out and documented.

Three common misconceptions

"Our chatbot is not a high-risk system, so the AI Act does not affect us."

The transparency obligation for directly interacting AI systems is not limited to high-risk systems. An ordinary service chatbot can have a low or limited risk and still fall under Article 50.

"A bot symbol makes the AI obvious."

Whether the "obvious" exception applies depends on the context and the perception of a discerning person. Clear text is more robust than an interpretation of colors, avatars, or animations.

"We put the notice in the privacy policy."

A privacy policy is important but does not automatically fulfill the requirement to inform users clearly at the latest during the first interaction. The notice belongs directly at the chat contact point.

A lean audit process for the coming weeks

1. Inventory: On which websites and in which languages is the chatbot active?
2. Visual check: Is "AI" clearly recognizable before or during the first message?
3. Check mobile and accessibility: Do the notice, focus, and contrast also work on small displays and with assistive technologies?
4. Test escalation: What happens if the bot does not have a reliable answer?
5. Control data protection separately: Which data are processed and how are users informed?
6. Secure evidence: Document texts, screenshots, responsibilities, and audit dates.
7. Schedule review: Test again after product, model, or design changes.

Conclusion: A clear AI notice is small but decisive

For most company websites, the technical change is manageable: a clear designation, a transparent greeting, accessible presentation, and a sensible path to human help. The organizational part is equally important: responsibilities, knowledge maintenance, and regular checks must be clarified.

Those who check these points before August 2, 2026, not only improve their regulatory starting position. A visibly honest AI chatbot creates trust and sets realistic expectations – two prerequisites for better conversations, more qualified inquiries, and less frustration in support.

Official sources and status

Status of this article: July 10, 2026. The EU guidelines on Article 50 were available as a draft at the time of editing.

• Regulation (EU) 2024/1689, especially Article 50
• EU Commission: Overview, risk classes and timeline of the AI Act
• Draft guidelines for the implementation of Article 50 from May 8, 2026
• Code of practice on transparency for AI-generated content from June 10, 2026
• Assessment of the EU Commission from July 9, 2026