Do I legally have to tell users they are talking to a chatbot?

Yes, if you have EU users. Article 50 of the EU AI Act requires providers of AI systems intended to interact with natural persons to design the system so users are informed they are interacting with AI, unless this is obvious from the circumstances.

Is a disclosure in the Terms of Service enough?

No. The disclosure needs to be clear and timely at the point of interaction, not buried in a legal document most users never read. A visible label like "You are chatting with an AI assistant" at the start of the conversation is the accepted approach.

Does a recommendation engine need the same disclosure as a chatbot?

Recommendation engines are typically limited-risk under Article 50 too, but the disclosure obligation is more directly about content generation and direct interaction. If your recommender only ranks products without generating text or interacting conversationally, the strongest requirement is usually the general obligation to avoid deceptive design, review your specific case with the free assessment.

What if my chatbot also screens job candidates?

A chatbot that filters or scores candidates for employment moves out of the limited-risk Article 50 regime and into the Annex III high-risk category, which brings much heavier obligations: technical documentation, human oversight, and a conformity assessment.

Limited risk, Article 50

EU AI Act rules for chatbots and AI assistants

If your product talks to EU users, whether it is a support bot, a shopping assistant, or a recommendation engine, Article 50 already applies to you. Here is exactly what it requires.

The rule

What Article 50 actually requires

Article 50 covers AI systems designed to interact directly with people. The core rule is simple: users must know they are talking to AI, unless it is obvious from the context (like an obviously robotic IVR menu).

This applies to most conversational products on the market today: customer support chatbots, sales assistants, onboarding bots, and AI-powered search or recommendation tools that generate natural-language responses. It has applied since the transparency obligations for general-purpose AI took effect in August 2025.

In practice

What compliant disclosure looks like

Do this

Show a visible label at the start of the conversation ("You are chatting with an AI assistant")

Keep the disclosure visible throughout the interaction, not just on first load

Label AI-generated images, audio, or video where technically feasible

Re-disclose after a human handoff back to AI (and vice versa)

Avoid this

Bury the disclosure only in your Terms of Service or Privacy Policy

Use a chatbot persona name (like "Sarah") without any AI indicator nearby

Rely on users to "figure it out" from response speed or phrasing

Assume disclosure is unnecessary because the bot only handles FAQs

Watch out for scope creep

When a chatbot stops being limited-risk

The classification follows what the system does, not what you call it. The same underlying chatbot technology can jump from limited-risk to high-risk the moment it starts doing any of the following:

Screening or ranking candidates for a job

Approving or denying a credit application, insurance quote, or discount tier tied to a credit-like decision

Triaging or recommending a medical or health-related course of action

Determining access to essential public or private services

Not sure which tier your chatbot falls into?

Answer 7 questions and get a definitive classification in about 3 minutes.

Start the assessment

Risk Levels Guide

The 4 EU AI Act risk levels explained, with examples

See how prohibited, high, and minimal risk compare