How to disclose chatbots and AI agents under the AI Act: wording and placement

KEY TAKEAWAYS Summary by the editors

1. Article 50(1) of the AI Act requires AI systems that interact directly with people to be designed so that those people know they are dealing with a machine, unless this is obvious to a reasonably well-informed person.
2. The information must be provided at the latest at the time of first interaction or exposure, in a clear and distinguishable way, and must meet applicable accessibility requirements.
3. The duty applies from 2 August 2026, and breaches of Article 50 can attract fines of up to EUR 15 million or 3 per cent of worldwide annual turnover, whichever is higher.
4. The disclosure is a legal minimum, not a trust strategy: a short, plain statement plus a clear route to a human tends to be easier to defend than a vague label.
5. Deepfake and synthetic-content rules in Article 50 are separate from chatbot disclosure, so product imagery and virtual models need their own review.

Under Article 50(1) of the EU AI Act, a chatbot or AI agent must be designed so that the person using it knows they are dealing with a machine, unless that is obvious to a reasonably well-informed person. The information has to arrive at the latest at the first interaction, in a clear and distinguishable form. In practice, that means a visible label and a short plain-language sentence at the start of the conversation, not a line buried in the terms and conditions.

What does the AI Act require when a customer talks to a chatbot?

Article 50(1) places the design duty on providers of systems that interact directly with natural persons: they must ensure the people concerned are informed that they are interacting with an AI system, unless it is obvious from the viewpoint of a reasonably well-informed person. Article 50(5) adds the form: the information must be given at the latest at the time of first interaction or exposure, presented in a clear and distinguishable manner, and must meet applicable accessibility requirements.

The obligations began to apply on 2 August 2026. Non-compliance with Article 50 falls under the tier in Article 99 of up to EUR 15 million or 3 per cent of worldwide annual turnover, whichever is higher, with the lower of the two figures applying to small and medium-sized enterprises.

Who is responsible, the retailer or the chatbot vendor?

The duty to design the system sits with the provider. A company that licenses a chatbot is usually a deployer, but a brand that has an assistant built and launches it under its own name or trademark can itself be the provider. The simplest route is to treat disclosure as a shared task. Ask the vendor to confirm the system supports the disclosure, then verify in the live customer journey that it appears, since it is the retailer's storefront on which the customer sees it.

The Act excepts cases where it is obvious, from the point of view of a reasonably well-informed person, that they are interacting with an AI system. A feature branded and presented as an AI assistant with a robot icon may qualify. A chat window named after a human-sounding stylist, a live-chat widget with an avatar photo, or a voice agent on a customer hotline does not make this obvious. The sensible approach for fashion retail is to disclose by default and not rely on the exception, because the test is judged from the customer's side and is not a settled matter.

Read also

AI in fashion: the September 2026 review

What wording works for an AI disclosure?

The Act does not prescribe wording. Good wording is short, specific and places the machine status ahead of any other claim. Examples of structure, to be adapted to brand voice and checked by counsel:

• "You are chatting with an AI assistant, not a person. Ask for a colleague at any time."
• "This is an automated assistant. It can help with sizing, orders and returns. For anything else we will connect you to our team."
• For voice: "This is an automated assistant from [brand]. You can say 'agent' to speak to a person."

Avoid human names without an AI marker, photographs of people as avatars, phrases such as "our stylist Anna" without qualification, and typing animations designed to imitate a human operator. These choices work against the purpose of Article 50(1) even where a label exists somewhere on the page.

Where should the disclosure be placed?

The timing rule is the anchor: at the latest at the first interaction. That points to three placements that can be combined: a persistent label on the chat launcher or window header, the opening message of every new session, and, for voice, the first sentence spoken. A cookie-style banner, a footer link or a clause in privacy policy text does not meet the clear and distinguishable standard by itself. Accessibility matters as well, so the label should be available to screen readers, readable at small sizes and offered in the language of the conversation.

How is this different from labelling AI-generated images and text?

Chatbot disclosure under Article 50(1) is about interaction. Marking and labelling of generated content is handled elsewhere in the same article. Providers of systems that generate synthetic audio, image, video or text must mark outputs in a machine-readable format. Deployers that publish deepfakes must disclose that the content is artificially generated or manipulated, with lighter disclosure for evidently artistic, creative, satirical or fictional works. A retailer using virtual models in campaigns therefore faces a second question, on top of chatbot labelling.

The Commission has published a voluntary Code of Practice on Transparency of AI-generated Content to help providers and deployers meet Article 50(2), (4) and (5). It is not mandatory, but the Commission and the AI Board have confirmed it is an adequate voluntary way to demonstrate compliance, and the EU has created a set of icons for deployers to label AI-generated content. Reported signatories numbered about 190 organisations by the end of July 2026.

Read also

What is agentic commerce? How AI agents will shop for fashion customers

How can a team test that the disclosure works?

A short test routine catches most gaps:

1. Open the storefront on a phone and on a desktop as a first-time visitor and check the label appears before you type.
2. Start the chat through each entry point: product page, order status page, messaging app and voice line.
3. Read the first reply as a customer would and confirm it states the assistant is automated.
4. Ask the assistant directly whether it is a human and check that it answers truthfully.
5. Confirm a route to a human exists and works during stated opening hours.
6. Keep screenshots and the wording version in a log with dates.

As shopping agents begin to contact retailer systems for customers, two situations arise. Your own assistants that talk to customers must disclose. Third-party agents that call your services are, in most cases, the responsibility of their provider and deployer. For fashion businesses, a practical step is to log automated traffic, state clearly in your interfaces when replies are machine-generated, and keep human escalation available for orders, returns and complaints.

Frequently asked questions

Do I have to tell customers a chatbot is AI under the AI Act?

Yes, in most cases. Article 50(1) requires AI systems that interact directly with people to be designed so those people are informed they are dealing with a machine, unless that is obvious to a reasonably well-informed person. The information must come at the latest at first interaction.

When do the chatbot disclosure rules start?

Article 50 transparency obligations began to apply on 2 August 2026. Providers of generative systems placed on the market before that date have until 2 December 2026 for the separate machine-readable marking duty in Article 50(2).

Is a small line in the terms and conditions enough?

It is unlikely to be. Article 50(5) requires information to be clear and distinguishable and given at the latest at first interaction, so a visible label and an opening message are a safer approach than a clause in legal text.

What are the penalties for missing the disclosure?

Breaches of Article 50 fall under the tier of up to EUR 15 million or 3 per cent of worldwide annual turnover, whichever is higher. For small and medium-sized enterprises, the lower of the two amounts applies.

GuideThe complete guide to AI in fashion e-commerce, marketing and retailRead the complete guide