AI Act Article 50: what a chatbot disclosure needs to explain

For a customer-facing chatbot, the first practical Article 50 question is whether people receive the required information that they are interacting with AI. That interface question is separate from the provision's rules about marking and disclosing generated content. A single badge should not be treated as proof that every relevant obligation has been met.

This guide helps a buyer scope the interface and the evidence needed for review. It does not determine all obligations for a particular service. Sources were checked on 11 October 2026.

Different paragraphs address different tasks

Article 50(1) concerns providers of systems intended to interact directly with people, with an exception where the AI interaction is obvious under its stated test. Paragraph 2 concerns providers' machine-readable marking of synthetic outputs. Paragraph 4 addresses deployer disclosures for deepfakes and specified public-interest text, with conditions and exceptions. Paragraph 5 requires relevant information to be clear, distinguishable and accessible, no later than first interaction or exposure. See the current Article 50 text.

Map the relevant paragraphs to the actual product and organisations involved. A company publishing a chatbot and a company using generated material in a public communication may have different implementation work to commission. Ask the reviewer to identify which requirements apply before treating an interface checklist as complete.

A fictional shop assistant looks like a staff conversation

Imagine a shop adding a chat window named “Customer service”. The first message invites the customer to ask a question, but nothing in the visible exchange explains that replies come from AI. The only disclosure is in a policy page linked from the footer.

A proposed design would put the information in the conversation entry point, where the customer can encounter it before relying on the reply. The exact wording and placement need review in context. A small icon with no explanation may not communicate what the project team assumes it does.

The same design should explain how to reach a person when the assistant cannot help. That is a proposed service choice, rather than a claim that Article 50 requires one particular handover mechanism. If the chat says a request was passed to staff, the system needs evidence that the handover actually happened.

Check the real entry points

A buyer should ask to see the chatbot on mobile and desktop, including a conversation opened through a direct link. A notice that appears on the home page may be absent when someone arrives elsewhere. Language selection and returning visits can also change what is visible.

The acceptance record can identify the reviewed wording, the places it appears and the tested version. If the team later embeds the assistant in another channel, someone needs to check that channel too. A screenshot of one initial design is not evidence for every future integration.

Output handling deserves a separate question. If the service allows a generated file to be downloaded or shared, establish what marking the supplier provides and what survives the delivery path. Do not assume a visible chat notice answers that technical question.

What to commission

Syntalith can scope the disclosure and handover requirements, implement the agreed interface and test the identified entry points. Your team supplies the intended audience, channels and the specialist assessment of applicable obligations. The proposal should distinguish interface work from content-marking integration and legal review.

For the fictional shop, the initial scope could cover the web widget and its two language versions, with any later messaging integration priced separately. Before agreeing it, we would inspect the existing chat entry points and the supplier's available settings. An existing configuration may already support the required design.

Describe where customers will encounter the assistant. You can consult service pricing without preparing a full technical specification. For the broader rollout, start with the provider and deployer roles and confirm the applicable timeline with your adviser.