Does my chatbot need an AI disclosure under the EU AI Act?

Article 50(1) covers AI systems intended to interact directly with natural persons. Providers must make sure people are told they are interacting with an AI, unless that is obvious (see the exception below). A narrow law-enforcement exception does not cover website chatbots.

Article 50(5) sets how that information is given: "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure". Article 50 "does not impose distinct or additional accessibility requirements"; existing accessibility rules apply where they bind you (Commission Guidelines ¶144). As good practice (not a separate Article 50 rule), write the disclosure in the same language as the chat and make sure screen readers announce it, so every visitor can notice it. A footer, privacy page, or terms clause alone is unlikely to meet that standard for a live chat.

For placement, edge cases and avatars, see the hub: Article 50 chatbot disclosure.

The Commission's Guidelines on transparency obligations were adopted on 5 November 2026 as C(2026) 7702 final. They are non-binding (paragraph 5). The 20 July 2026 text, C(2026) 5054 final, was a draft submitted for approval. The adopted text sets a practical bar for a chatbot disclosure:

• One notice at the start, in most cases (¶40): a single prominent notification before the first interaction is likely to suffice in most instances. Riskier or longer immersive interactions, sensitive advice, vulnerable users and AI companions may need periodic or context-aware reminders. The guidelines set no universal interval. This is the same substantive advice as the July draft, not a new November obligation.
• First turn, with persistent labels as appropriate (¶37): first-turn greetings "may be combined, as appropriate, with persistent badges or labels visible throughout the interaction". A persistent label is suggested where appropriate, not required.
• Generic notices aren't enough (¶38): "a general disclosure like 'Services on this website use AI' is insufficient", and a disclosure only in terms and conditions isn't enough on its own.
• Clear means noticeable (¶142): the guidelines consider information clear where it is "noticeable, easy to understand by and accessible for the natural person concerned (including persons with disabilities)".

Put the disclosure in the chat itself, from the first message. A site-wide "we use AI" line is not enough on its own. One prominent in-chat notice is likely to be enough in most cases; some riskier chats need reminders.

Article 50(1) is written for the provider: the party that develops the chatbot (or has it developed) and places it on the market or into service under its own name or trademark. Many site owners are deployers of a vendor widget (Intercom, Zendesk, Drift, Crisp, and similar).

Deployers have their own duties under 50(4) (deepfakes, public-interest AI text). For a chatbot, though, you control what visitors see, so check the vendor's disclosure settings, then add a clear greeting yourself.

Vendor defaults change. Intercom's own documentation says the Fin "AI Agent" label is on by default only for workspaces that used Fin before a cut-off in July 2025, and off by default for newer ones (Intercom Help: AI Agent disclosure). Step-by-step settings: how to add disclosure for Intercom, Zendesk, Drift and Crisp.

You can skip the disclosure only where it is obvious to a "reasonably well-informed, observant and circumspect" person that they are dealing with an AI (Art. 50(1)). Treat that exception as narrow. Many chat widgets use human names and avatars, and the same window often switches between a bot and a human agent — so "it looks like a bot" is rarely enough on its own.

Put a clear notice in or directly above the chat window so visitors see it before the first message. One line you can adapt:

AI-powered chat · Replies are written by an AI system and may contain mistakes.

More variants, HTML banners, and placement notes: AI chatbot disclaimer template.

The Digital Omnibus did not delay the chatbot disclosure duty in Article 50(1), which has applied since 2 August 2026.

The Omnibus gave a short grace period for machine-readable marking only, of AI-generated content from systems placed on the market before 2 August 2026. 2 December 2026 is not "the Article 50 deadline". Full timeline: EU AI Act Article 50 delay.

Breaches of Article 50 can be fined up to €15 million or 3% of worldwide annual turnover, whichever is higher (the lower of the two for SMEs), under Art. 99(4)(g) and 99(6), which covers "transparency obligations for providers and deployers pursuant to Article 50" (AI Act Service Desk, Article 99). What decides the amount: what happens if you ignore Article 50.

Most chatbot disclosure failures come from where and when the notice appears, not from its wording.

• Footer or terms-only notice. A disclosure only in terms and conditions is insufficient on its own (Guidelines ¶38).
• A generic "this site uses AI" banner. Not enough on its own (¶38).
• A human name and avatar with no AI label. The "obvious" exception is narrow (Art. 50(1)).
• Assuming the vendor's label is on. Defaults change; check the widget as a visitor sees it.

For the wider list of Article 50 checks, use the Article 50 compliance checklist.

Our State of Article 50 study found 58 of 59 popular EU websites with a chat widget showed no AI disclosure detected by an automated scan at first view.

1. Open your site as a new visitor — use a private window so logged-in agent views don't hide the visitor greeting.
2. Read the first chat message — does it say, in plain language, that this is an AI (for example "I'm an AI assistant")?
3. Ask "Am I talking to a person?" — confirm the reply states it is an AI. Keep a dated screenshot or transcript.

Scan for free — the free Article50.io scan opens your live page in a real browser and looks for a clear AI disclosure. What it checks and what it can't: how the scan works.