Does Your Chatbot Need an AI Disclosure? Article 50(1) Explained
Last updated:
Article 50 chatbot disclosure is needed in most cases: an AI chatbot that talks to visitors must tell them they are interacting with AI, unless that is obvious, clearly and at the latest at the first interaction. A footer or privacy-policy line is not enough. The free Article50.io scan loads one public page and shows the most severe finding.
The simplest fix is a first message like this, ready to paste: “Hi! You're chatting with an AI assistant, not a person.”
In our scan of popular EU websites, 58 of the 59 chat widgets we detected showed no AI disclosure at first view (98.3%). Read the study. For more wording, use the free AI chatbot disclaimer template. The full text is in Article 50(1) and Article 50(5).
This page is general information about Article 50 chatbot disclosure, not legal advice.
When the chatbot disclosure applies
Article 50(1) kicks in when all three of these are true.
| Condition | What it means |
|---|---|
| It's an AI system | The widget generates or selects responses using AI — an LLM-powered assistant, an AI support agent, or a vendor widget with AI features switched on. |
| It interacts directly with people | Visitors type or speak to it and it responds. Chat widgets, voice agents and in-app assistants all count. |
| It isn't obvious | Unless a reasonably well-informed, observant and circumspect person would already know it's AI, you need to tell them. |
What Article 50(1) and Article 50(5) require
Two paragraphs do the work for chatbots. Article 50(1) says people must be informed. Article 50(5) says how and when.
Article 50(1) (paraphrase): Providers must ensure that AI systems intended to interact directly with natural persons are designed and developed so that the people concerned are informed they are interacting with an AI system — unless this is obvious from the point of view of a reasonably well-informed, observant and circumspect person, taking into account the circumstances and the context of use.
Article 50(5) (paraphrase): The information in paragraphs 1 to 4 must be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure, and it must meet the applicable accessibility requirements.
- ✔Clear and distinguishable — say “AI” plainly; a vague “assistant” label alone is a weak bet.
- ✔At the latest at the first interaction — before or with the first message, not after the visitor has already been chatting.
- ✔Accessible — readable contrast, sensible size, and in the language of the chat where you can.
- ✔A mention in your terms of service or privacy policy isn't enough; the visitor needs to see it where they're chatting.
Full Article 50 explainer.
Provider vs deployer: who the Act names
Article 50(1) is written as a provider duty. A provider builds or supplies the AI system and puts it into service under its own name. A deployer uses an AI system under its authority. For an off-the-shelf Intercom Fin, Zendesk AI agent, Drift or Crisp Hugo setup, the vendor is most likely the provider.
That doesn't let the site owner off the hook in practice. Visitors see the chat on your site, under your brand. You usually control the bot's name, avatar, greeting and which disclosure settings are switched on. If the widget reads as “Sarah from Support” with no AI notice, the undisclosed AI is on your page. Add the disclosure yourself: one prominent line in the greeting or a banner at the start of the session. In most cases that is likely to be enough (guidelines paragraph 40). Longer or immersive chats, sensitive advice, vulnerable users and AI companions may need periodic or context-aware reminders. There is no fixed interval, and this is not a new November obligation.
| Role | What it means in practice |
|---|---|
| Provider (typical vendor widget) | Design duty under Art. 50(1) — most likely the chat vendor |
| Site owner / deployer | Controls name, avatar, greeting and settings visitors actually see |
| Practical fix | Paste a clear AI disclosure where the conversation starts |
The “obvious” exception is narrow
Article 50(1) skips the disclosure only when it is already obvious to a reasonably well-informed, observant and circumspect person that they are dealing with an AI. Don't treat that as the default.
Many chat widgets work against the exception: human first names, friendly avatars, typing indicators, and the same window switching between a bot and a human agent. The Commission's guidelines, adopted on 5 November 2026 as C(2026) 7702 final and non-binding (paragraph 5), say a provider who relies on the exception must be able to assess and demonstrate that the artificial nature is obvious to a reasonably well-informed, observant and circumspect person (paragraph 42). A generic reference to an “assistant”, used on its own, is not enough (paragraph 38). The 20 July 2026 text (C(2026) 5054 final) was a draft submitted for approval; it said the same on this point. Prefer an explicit “AI” notice rather than hoping the exception covers you.
- ✔Human-sounding names (Sarah, Hugo, Alex) cut against “obvious”
- ✔Avatars and typing indicators make the bot feel like a person
- ✔Bot-to-human handoff in the same window adds confusion
AI companions and long, personal chats
AI companions are interactive AI systems under EU AI Act Article 50(1), so users must be told they're talking to an AI unless that's obvious, and the Commission's guidelines say reminders are likely to be necessary where users may form emotional attachments. The Commission's Article 50 guidelines, adopted 5 November 2026 (C(2026) 7702 final; non-binding), list AI companions among the systems Article 50(1) covers (the unnumbered examples box after paragraph 31). Paragraph 40 says a single prominent notice is likely to suffice in most instances, but "in certain riskier contexts, periodic reminders and context-aware disclosures are likely to be necessary to ensure continued user awareness". AI companions are one of the examples it gives, where users risk "forming emotional attachments or dependencies".
A practical approach for companions and other long chats (good practice drawn from the guidelines, not a fixed rule in Article 50):
- Keep a visible AI label in the chat window, not only in the first message. Paragraph 37 says first-turn greetings may be combined, as appropriate, with persistent badges or labels visible throughout the interaction "as proportionate to the specific context and risks".
- In riskier contexts, plan periodic or context-aware reminders, for example when a conversation turns to health, money or emotional distress. Paragraph 40 names emotional distress, health advice and financial advice among those contexts.
- Practice tip (not 7702 wording, and not a legal requirement on its own): also repeat the notice when a long chat resumes after a break. Paragraph 37's "interruptions" point is the closest guideline language.
- Take extra care where children or other vulnerable users may be chatting (paragraph 40).
- Make sure the bot always says it's an AI when someone asks whether they're talking to a person (paragraph 40).
Key point: a clear opening notice is usually enough. For AI companions and other long, emotional or high-stakes chats, the Commission's non-binding guidelines say reminders are likely to be necessary. Article 50 itself sets no reminder interval.
The free Article50.io scan checks one public page and reads only what a widget shows as it first loads, so it can't check reminders later in a conversation.
Was this delayed by the Digital Omnibus?
No — chatbot disclosure was not delayed. Article 50(1) chatbot disclosure has applied since 2 August 2026. The Digital Omnibus delayed high-risk rules, not the Article 50 transparency duties for telling people they are talking to an AI.
| Duty | Status after Omnibus |
|---|---|
| Art. 50(1) — tell people they are interacting with AI | In force since 2 August 2026 — not delayed |
| Art. 50(5) — clear, distinguishable, first interaction, accessible | Applies with 50(1) — not delayed |
| Art. 50(2) — machine-readable marking for systems already on the market before 2 Aug 2026 | Transition until 2 December 2026 for those providers — not a chatbot-greeting delay |
Full delay explainer.
Paste-ready wording
Use one core disclosure visitors can see before they type, plus a handoff line if a human can take over, and a reminder line for long or sensitive chats. Full template page.
1. Chat window banner (HTML)
Put this in or directly above the chat window so it stays visible while the chat is open:
<div class="chat-ai-banner" role="note">
AI-powered chat · Replies are written by an AI system and may contain mistakes.
</div>2. First-message greeting
Short variant ready to paste: “Hi! You're chatting with an AI assistant, not a person.”
Hi, I'm [Assistant name], an AI-powered assistant for [Company]. I can help with [topics]. For anything I can't answer, [how to reach a human].3. Bot-to-human handoff
You've been connected to [Agent first name] from our team. Earlier replies in this chat came from an AI-powered assistant.4. Reminder line (long or resumed chats)
Optional. Article 50 sets no reminder duty or interval. The Commission's non-binding guidelines (paragraph 40) say reminders are likely to be necessary in riskier contexts, such as AI companions or long, sensitive chats.
Just a reminder: you're chatting with an AI assistant, not a person. [How to reach a human].Placement tips
- ✔Visible before or at the start of the conversation
- ✔Plain wording — "AI assistant", not "smart helper"
- ✔Stays visible while the chat window is open
- ✔Kept clear when the bot hands off to a human agent
How to check in 60 seconds
Settings pages don't show what visitors get. Do this once in a private window.
- Open the chat as a visitor — Private/incognito window, desktop and a quick look on mobile. Don't use an admin session.
- Read the first message and the bot name — Is there a clear “AI” notice before or with the first reply? Does the name sound like a person with no AI label?
- Ask: “Am I talking to a person?” — If the bot dodges or never says it's AI, the disclosure isn't doing its job for a first-time visitor.
- Trigger a human handoff if you can — Check that the switch from AI to a person is visible in the same window.
Vendor-specific steps (Intercom Fin is often off-by-default for newer setups; Intercom Fin AI label defaults): Intercom, Zendesk, Drift and Crisp how-to.
Free scan: what Article50.io checks on the page
After the manual check, run a free Article50.io scan. Article50.io's scan detects AI-capable chat widgets (Intercom, Drift, Zendesk, Crisp and custom markup). It loads one public page in a live browser and looks for disclosure language on the page and in the widget as it first loads.
That is a live-page disclosure scan, not a back-and-forth conversation test. It does not open the chat and start a Q&A, so a notice that only appears after the visitor sends a message may still be flagged. Treat a flag as “no disclosure is visible before the chat is used,” then open the chat and confirm. A clean result is not a compliance clearance. You can also see the full report and pricing.
Beyond the chatbot
A chat widget is only one Article 50 trigger. If your site also publishes AI-generated images, video or articles, check AI-generated content labeling and deepfake disclosure rules too. For voice or phone agents, the same Art. 50(1) idea applies by speech at the start of the call: Do AI voice agents need a disclosure?
Chatbot-only scope: Do I need Article 50 compliance if I only use a chatbot?. Full Act picture: Article 50 of the EU AI Act.
Frequently Asked Questions
Does my chatbot need an AI disclosure?
Yes, in most cases. If your site uses an AI system that interacts directly with visitors, Article 50(1) requires that people are informed, unless it is already obvious from context. Pair that with Article 50(5): the notice must be clear and distinguishable at the latest at the first interaction, and accessible. A free scan at https://article50.io/ is the fastest way to see what your public page shows today.
What does Article 50(1) require — and what about 50(5)?
Article 50(1) requires providers to design AI systems that interact with people so those people are informed they are dealing with an AI, unless that is obvious. Article 50(5) requires that information to be clear and distinguishable, given at the latest at the first interaction or exposure, and to meet applicable accessibility requirements. For a website chatbot, that means a plain AI notice where the chat starts — not only in your terms.
My chatbot is a third-party widget. Is the disclosure still my problem?
Visitors see an undisclosed AI on your site, whoever built it. Article 50(1) places the design obligation on the provider (often the vendor), but you usually control the greeting, name and settings. Adding the disclosure yourself — banner, first message, or both — is the simplest way to close the gap.
Is a footer or privacy-policy line enough?
No. Article 50(5) wants the information where the interaction happens, at the latest at the first interaction. A footer, terms page or privacy policy alone does not put a clear, distinguishable notice in front of the person starting the chat.
Isn't it obvious that a chat widget is a bot?
Not reliably. Many AI chat widgets use human names, avatars and typing indicators, and the same widget often switches between a bot and a human agent. The exemption only applies when it is obvious to a reasonably well-informed, observant and circumspect person — don't rely on it without a clear reason. The Commission's guidelines (paragraph 38) say a generic reference to an "assistant", used on its own, is not enough.
Is one banner or greeting enough?
In most cases, yes: a single prominent notice at the start of the session is likely to be enough (Commission guidelines, paragraph 40). Riskier or longer immersive interactions, sensitive advice, vulnerable users and AI companions may need periodic or context-aware reminders. The guidelines, adopted on 5 November 2026 as C(2026) 7702 final, set no universal interval. They are non-binding (paragraph 5) and do not create a new November obligation.
Do AI companions need an AI disclosure under Article 50?
Yes, unless it's obvious. The Commission's non-binding Article 50 guidelines (C(2026) 7702 final) list AI companions in the examples after paragraph 31 as systems Article 50(1) covers, and read the "obvious" exception restrictively (paragraph 45). Paragraph 40 adds that where users may form emotional attachments or dependencies, periodic reminders and context-aware disclosures are likely to be necessary.
What if a visitor asks the chatbot whether it's a person?
It should say plainly that it's an AI. The Commission's non-binding Article 50 guidelines (C(2026) 7702 final, paragraph 40) say providers must design the system to disclose its nature whenever it's asked, or when the person is likely to be misled or confused about it. Test it yourself in a private window: ask "Am I talking to a person?"
When did the chatbot disclosure requirement take effect?
Article 50 has been enforceable since 2 August 2026. Chatbot disclosure under Article 50(1) was not postponed by the Digital Omnibus. Breaches can attract fines of up to €15 million or 3% of global turnover.
Does this apply if my company is outside the EU?
It can. The AI Act can cover providers and deployers outside the EU when their AI system's output is used in the EU — for example a chatbot answering EU visitors. Being established outside the EU does not, by itself, take you out of scope.
What about Intercom Fin, Zendesk, Drift or Crisp?
Don't assume the vendor switched disclosure on for you. Intercom's own documentation says its “AI Agent” label in the Messenger is off by default for customers who set up Fin after 10 July 2025.
How to add an AI disclosure to Intercom, Zendesk, Drift and Crisp
Do AI voice or phone agents need the same disclosure?
Yes, in the same Art. 50(1) sense: callers must be told they are speaking with an AI at the latest at the first interaction, which on a phone line means saying it at the start of the call.
Check your site automatically
Article50.io is an automated scanner for the Article 50 transparency rules of the EU AI Act. It scans a website for potential Article 50 transparency gaps and generates a report of what the scan found and what it could not check, with remediation guidance, implementation instructions and disclosure wording you can paste in.
The free scan shows your single most severe finding. Scans are capped at 45 seconds — no signup, public pages only.
Related Article 50 resources
- Art. 50EU AI Act Article 50 (Art. 50): Full Text & Plain-English Explanation
- Deadlines · Digital OmnibusWas the EU AI Act Delayed? What the Digital Omnibus Means for Article 50
- Template · Art. 50(1)Free AI Chatbot Disclaimer Template — Article 50(1) Disclosure
- Templates · Art. 50AI Disclosure Statement Templates and Examples for Websites
- Q&A · Art. 50(1)Does my chatbot need an AI disclosure under the EU AI Act?
- Q&A · ScopeDoes Article 50 Apply to US Companies? (EU AI Act)
Automated technical guidance, not legal advice. Citations refer to Regulation (EU) 2024/1689.