Do you have to tell visitors your chatbot is AI?

A chat widget on a shop or a service site that answers visitors' questions is the common case: a support bot, a sales assistant, a booking helper. This guide covers the rules that reach those bots in the EU and in the US states where we found a duty. It is general information, not legal advice, and we checked every source on October 21, 2026.

The EU and the US frame the duty differently, so a single "label your chatbot" rule would be wrong in both places. Start from our AI disclosure laws hub, the page on Article 50 of the EU AI Act and the United States page for the wider map. Our AI disclosure laws finder asks a few questions and shows which of these may apply to you. AI-made people in advertising fall under different rules, covered in our guide to synthetic performer ads, and labels on AI images are covered in alt text for AI-generated images.

The EU: the provider's duty and the business's exposure

Article 50(1) of Regulation (EU) 2024/1689 says: "Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use." The Commission's Article 50 FAQ (last updated July 24, 2026) describes it the same way: the duty is on providers, and its list of deployer duties covers emotion recognition, deep fakes and AI-generated text on matters of public interest, not chatbots.

The date has not moved. The Commission's Guidelines on Article 50 (C(2026) 5054 final, July 20, 2026, non-binding according to paragraph 5) say in paragraph 153 that "compliance with the disclosure obligation for AI systems directly interacting with natural persons must be ensured as of 2 August 2026". The grace period to December 2, 2026 that Regulation (EU) 2026/1744 gave older systems covers only the marking duty for generated content, not chatbot disclosure.

Who is the provider when you embed a vendor's chatbot?

Paragraph 11 of the Guidelines gives three anchors. A company that provides a chatbot "under its own name or trademark" to customers, including customers who integrate it into their processes "without the need for modifications or additional components", is the provider. A company that builds a chatbot in-house and puts it into service under its own name is a provider. A company that takes an existing generative system, modifies it, for example with new training data, and puts it into service under its own name becomes the provider of the new system.

On our reading, a store that embeds an unmodified vendor widget is a deployer and the vendor carries Article 50(1). The Guidelines do not address white-label or commissioned bots that a business brands as its own, and Article 3(3) of the Act also covers whoever has a system developed and puts it into service under their own name. That gray zone is a question for a lawyer and for your vendor contract.

A deployer still has exposure. You choose the configuration, and paragraph 38 says "human-like representations that may mislead users" do not work as a disclosure. Paragraph 50 adds that consumer law runs in parallel: where a service is AI-driven, the AI functionality "may qualify as an essential characteristic that must be disclosed before the consumer is bound by a contract".

What counts as a disclosure in the EU

• In the chat, before the first exchange. Paragraph 40 says "a single, prominent notification before the first interaction" is likely to suffice in most instances, with reminders in riskier contexts such as financial advice, insurance, legal assistance, health advice and complaints handling, and for vulnerable users.
• Not only in the small print. Paragraph 38 lists what is insufficient alone: terms and conditions, URLs or documentation; metadata that people cannot perceive; unclear signals such as a generic "assistant"; and generalized statements. Its example is that, on a platform offering many services, a general disclosure like "Services on this website use AI" is insufficient.
• Answer when asked. Paragraph 40 says disclosure must happen "in all situations where the AI system is being asked questions relating to its nature".
• Not obvious. The exception for obvious cases is read restrictively. Paragraph 45 lists "AI chatbots embedded in online platforms or assistance support tools (helpdesks)" whose replies users "may perceive as human-generated" among the cases where the duty applies.
• Accessible. Paragraph 34 says the information "must also be provided in an accessible format for natural persons with disabilities". Our guide to accessible AI labels covers what that means for a label.

The Commission's FAQ says "Fines can reach up to 15 million euros or 3% of total worldwide turnover for the preceding financial year", with the lower of the two amounts applying to small and medium enterprises and small mid-cap companies (Article 99(6) and (6a)). We found no enforcement action under Article 50 anywhere as of October 21, 2026.

The US: duties on the business that uses the bot

In the US the duty is on the person using the bot, which for a shop's chat widget is the shop. We found no federal AI labeling rule that applies to ordinary chatbots, and the FTC's general authority over deceptive practices is separate. The state laws are narrow, and each has its own trigger.

Three limits matter. California SB 1001 needs an intent to mislead and a sales or voting purpose, so a clear label is the safe harbor. Utah's definition covers generative AI and leaves out scripted bots, while California's "bot" and Maine's "any other computer technology" are not limited to AI. And Texas's Responsible Artificial Intelligence Governance Act, in force since January 1, 2026, puts its disclosure duties on government agencies and health care providers and none on other private businesses. A wave of 2026 companion chatbot laws in other states mostly targets companion products, not customer service bots, so read the definitions in your state before assuming you are in or out.

What a good chatbot disclosure looks like

The statutes and the Guidelines point the same way. This is our summary, not a legal standard.

• Say it in the chat, at the start, in plain words. Paragraph 37 of the Guidelines gives prominent, plain-language labels or banners, such as "You are interacting with an AI system", and first-turn greetings in chatbots as examples, placed near the input field and in simple wording.
• Keep it visible. Utah's safe harbor asks for disclosure "throughout the interaction", and paragraph 37 suggests "persistent badges or labels visible throughout the interaction".
• Answer truthfully when asked. Set the bot's instructions so that "are you a real person?" gets a plain answer.
• Do not dress it as a person. A human name and photo can defeat a label. Paragraph 45 names a chatbot's human-looking profile picture as a factor in whether the interaction is obvious.
• Offer a person. AB 1609 requires a hand-off for large businesses, and it is good practice for everyone else.

A few wordings that fit what the sources above ask for:

Questions to put to your chatbot vendor

• Which company is the provider under the EU AI Act, and does it say its widget meets Article 50(1)?
• Does the widget show an AI notice by default, and can your team turn it off? If so, leave it on.
• What does the bot say when a visitor asks whether it is a person?
• Can you set a name and avatar? Pick ones that do not suggest a human.
• If a person takes over the chat, does the visitor see that? Paragraph 30 of the Guidelines says the AI-generated outputs need disclosure unless they were "properly reviewed and sent by humans as the main interlocutors".
• Does the contract say who is responsible for each disclosure duty?

What a check can and cannot tell you

A tool can read what a page declares. Our free AI label checker looks for known chat widgets and for AI-disclosure wording on the page where you run it, and our accessible AI label generator has a chatbot preset that writes disclosure wording and markup with the text equivalent included. Neither can tell whether a bot is AI-driven or what it says to a given visitor. A widget that loads after the page, or inside its own frame, may put its wording outside the page's HTML, so a page can show no wording found while the widget does disclose. Absence of wording proves nothing, and no check says a chatbot meets any law.