AI marketing

Do You Have to Tell Customers They're Chatting With AI? Disclosure Rules in 2026

When the law says you must tell customers they're chatting with AI: the EU AI Act, California, Utah and Maine rules, Meta's policy, and wording that works.

A customer in Lisbon sends your WhatsApp a question at 11pm: does the blue dress come in a 38? Your AI assistant answers in Portuguese within seconds, with the price and the delivery fee. Since 2 August 2026, EU law says that customer must be told, from the first message, that they're talking to an AI system. Customers in Maine, California and Utah are covered by state laws of their own. And where no law requires it, Meta still recommends telling people.

So, do you have to disclose your AI chatbot? Often yes, depending on where your customers are, and it's the sensible default where no rule demands it. Below are the main rules as they stand in November 2026, with wording you can copy.

This is general information, not legal advice. If your business depends on getting this exactly right in a particular country, speak to a lawyer there.

Who has to disclose, when and how

RuleWho it coversWhen you must discloseHow
EU AI Act, Article 50(1) and 50(5) (applies from 2 August 2026)Providers of AI systems that interact directly with people, including providers outside the EU whose systems are used in the EU"At the latest at the time of the first interaction", unless it's obvious to a reasonably well-informed person that it's AI"In a clear and distinguishable manner", meeting accessibility requirements
California, Business and Professions Code §17941 (operative since 1 July 2019)Anyone using a bot to talk to a person in California onlineLiability applies when the bot hides that it's a bot to mislead someone into a purchase or sale, or to influence a vote; disclosing removes that liability"Clear, conspicuous, and reasonably designed" to tell people it's a bot
Utah, S.B. 226 (2025), Utah Code Title 13, Chapter 75 (from 7 May 2025)Businesses using generative AI with consumers; stricter for regulated professionsWhen the customer clearly asks whether they're dealing with a human or AI; at the start for "high-risk" interactions in regulated professionsA safe harbour protects you if the AI discloses clearly and conspicuously at the start and throughout
Maine, Public Law 2025, chapter 294 (signed June 2025)Anyone using an AI chatbot or other technology in trade with a consumerWhenever a reasonable consumer could be misled into thinking they're dealing with a humanNotify the consumer "in a clear and conspicuous manner" that they're not dealing with a human
Meta: Messenger and Instagram messaging policyBusinesses and apps using Messenger or Instagram messagingWhere the law requires it: at the start of a conversation, after a significant lapse of time, and when a chat moves from a person to automation. Recommended everywhere elseTell the person they're interacting with an automated service
WhatsApp Business Messaging Policy (updated 23 September 2026)Businesses on the WhatsApp Business PlatformNo separate disclosure line in the policy; the laws above still applyAutomation is allowed in the 24-hour window, with "prompt, clear, and direct escalation paths" to a person

Europe: Article 50 of the AI Act

Article 50(1) says providers "shall ensure that AI systems intended to interact directly with natural persons are designed and developed" so that "the natural persons concerned are informed that they are interacting with an AI system." Article 50(5) adds that this must be done "in a clear and distinguishable manner" and "at the latest at the time of the first interaction or exposure." The European Commission's guidance page states plainly: "These transparency obligations apply from 2 August 2026."

Four details matter for a business that answers customers with AI.

It reaches beyond Europe. Article 2 applies the Act to providers placing AI systems on the EU market "irrespective of whether those providers are established or located within the Union or in a third country", and to providers and deployers outside the EU "where the output produced by the AI system is used in the Union". A shop in Johannesburg whose assistant replies to a customer in Paris can be within the Act's reach, even though the shop has no office in Europe.

The duty is written for the provider. Article 50(1) puts the obligation on whoever provides the AI system, not directly on every business that switches one on. If you use a ready-made tool, it's the tool's job to build the disclosure in, and your job to check it does and not to switch it off for European customers. If you build your own chatbot, or put one into service under your own name, the provider duty may fall on you.

The "obvious" exception is narrow. Disclosure isn't needed where it's obvious from the context. According to Addleshaw Goddard's summary of the Commission's July 2026 guidelines, the guidelines treat an internal assistant used by "properly-trained, AI-literate staff" as obvious, and describe "AI chatbots on online platforms where users receive AI outputs they may perceive as human-generated" as not obvious. A sales assistant replying on WhatsApp in a friendly tone is very likely to fall in the second group.

The "digital omnibus" didn't delay chatbot disclosure. The EU's AI omnibus, published in the Official Journal on 24 July 2026, postponed several parts of the Act. On Article 50, the Future of Privacy Forum's analysis says: "While Article 50 becomes generally applicable on 2 August 2026, the AI Omnibus postpones Article 50(2)", the separate rule on marking AI-generated content, giving systems already on the market until 2 December 2026. The Commission's AI Act Service Desk, whose text reflects the consolidated version of 27 July 2026, marks only paragraph 7 of Article 50 as amended. Paragraph 1, the chatbot rule, applies from 2 August 2026 as originally planned. Under Article 99, breaches of Article 50 can bring fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher; for small and medium-sized businesses and start-ups, it's whichever is lower.

Article 50(2), the machine-readable marking of AI-generated images, audio, video and text, is a different subject. If that's what you're after, our guide to the "AI info" label on Facebook and Instagram covers it.

United States: three state laws, built differently

As of November 2026 there's no US federal law that requires businesses to disclose chatbots; bills have been introduced in Congress but none has passed. Three state laws are worth knowing.

California: don't use a bot to deceive

California's law, added by SB 1001 and operative since 1 July 2019, makes it unlawful "to use a bot to communicate or interact with another person in California online, with the intent to mislead the other person about its artificial identity" in order to deceive them and encourage a purchase or sale, or to influence a vote. Then comes the key line: "A person using a bot shall not be liable under this section if the person discloses that it is a bot." The disclosure must be "clear, conspicuous, and reasonably designed to inform persons with whom the bot communicates or interacts that it is a bot."

"Bot" means "an automated online account where all or substantially all of the actions or posts of that account are not the result of a person," and "online" includes social networks and apps. You'll sometimes read that the law only covers sites with 10 million US visitors. That threshold is in the definition of an "online platform", and section 17942 says the law doesn't impose a duty on the service providers of online platforms. It isn't a size limit on businesses that use bots.

Utah: answer honestly when asked

Utah's S.B. 226, in force since 7 May 2025, rewrote the general rule. A business using generative AI in a consumer transaction must disclose it "if the individual asks or otherwise prompts the supplier about whether artificial intelligence is being used", and that prompt must be "a clear and unambiguous request to determine whether the interaction is with a human or with artificial intelligence." Regulated professions face a stricter rule for "high-risk" interactions, such as those involving sensitive health or financial information or personalised advice people may rely on for important decisions: disclosure up front.

The law also offers a safe harbour from enforcement if the AI "clearly and conspicuously discloses" that it is AI at the outset of the interaction and throughout it. The state's Division of Consumer Protection can impose administrative fines of up to $2,500 for each violation.

Maine: don't let customers think it's a person

Maine's law, signed in June 2025, is the broadest of the three. A person "may not use an artificial intelligence chatbot or any other computer technology to engage in trade and commerce with a consumer in a manner that may mislead or deceive a reasonable consumer into believing that the consumer is engaging with a human being unless the consumer is notified in a clear and conspicuous manner that the consumer is not engaging with a human being." A breach counts as a breach of the Maine Unfair Trade Practices Act. No intent to deceive is needed: the test is what a reasonable customer might believe.

What Meta and WhatsApp require

Meta's Messenger and Instagram messaging policy turns the law into three moments: the start of a conversation, after a significant lapse of time, and when a chat moves from a person to automation. It names California and Germany as markets that need attention, and recommends telling people "even where not legally required". The full set of Meta's messaging rules is in Instagram DM automation: Meta's rules.

The WhatsApp Business Messaging Policy takes a different angle. It says "You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths", such as handing the chat to a person in the conversation, a phone number or an email address. It doesn't add its own disclosure wording, but the laws above apply on WhatsApp just as they do anywhere else.

Writing the disclosure: examples you can use

Put together, the rules point to the same habit: tell people at the start, in the chat itself, in plain words, and never deny it later. A line in your Instagram bio or your terms and conditions is easy to miss, and the EU and US rules all talk about disclosure the customer will actually see.

The first line

Good first lines are short, name the business, say it's automated, and offer a person:

"Hi! You're chatting with Lume Boutique's automated assistant — you can ask for a person any time. Yes, the blue dress comes in a 38: €49, plus €4.90 delivery. Shall I reserve one for you?"

Notice that the disclosure doesn't replace the answer. The customer asked about a dress; they get the disclosure and the price in the same message, so honesty costs no speed. (Our piece on speed to lead explains why that first answer matters so much.)

Lines that fall short:

  • "Hi, I'm Sarah! How can I help?" A human name with no mention of automation is exactly what Maine's "reasonable consumer" test is aimed at, and what California's law targets when the bot is used to push a sale.
  • "Powered by AI" in small text under a web chat window. Unlikely to be "clear and conspicuous" or "clear and distinguishable", and on WhatsApp there's nowhere to put it anyway.
  • A disclosure buried after three paragraphs of welcome text. Make it the first sentence.

Write it in the customer's language. A disclosure in English to a customer writing in French is hard to call clear.

When the customer asks "am I talking to a bot?"

This is the moment Utah's law is built around, and where a dishonest answer would be hard to defend anywhere. Answer directly, then offer a person:

"I'm the automated assistant here, and I can bring in someone from the team right away if you'd like."

Avoid dodges such as "I'm here to help you!" or "Ha, I get that a lot." They answer a different question and leave the customer more suspicious than before.

When a person takes over, and when the assistant takes back

A handover line tells the customer the next reply is from a person: "I'm passing you to Inês from our sales team; she'll reply shortly." Meta's policy also flags the reverse, a chat moving from a person back to automation. If your teammate steps away and the assistant carries on, a short "I'll hand you back to our assistant for now" from your teammate covers it.

Inside Rumidza: when the assistant says it's automated

Rumidza's AI replies to customers on WhatsApp, Facebook Messenger and Instagram in your business's name. This is exactly how disclosure works in the app today, taken from its code and its terms:

  • Every new conversation opens with one line. A conversation counts as new when the assistant hasn't replied to that customer in the past 30 days. The first reply begins with a short sentence, in the customer's language, saying they're chatting with your business's automated assistant and can ask for a person any time, followed by the answer to their question. If the AI's first reply doesn't mention an assistant, automation, a bot or AI, Rumidza puts a fixed line in English in front of it: "Hi! You're chatting with <your business>'s automated assistant — you can ask for a person any time."
  • After that, replies aren't labelled. The conversation reads naturally, and the assistant is told not to keep repeating that it's automated unless asked.
  • It's instructed never to claim to be human. If a customer asks whether they're talking to a person, a bot or an AI, the assistant is told to answer honestly, as in the example above, and if the customer wants a person it flags the chat for your team with the reason.
  • The first message is always in writing. If you've switched on voice-note replies, the opening message that says it's your automated assistant still goes as text.
  • You can switch the line off, with one exception. The setting is in Settings ("Start each new chat with one line saying it's your automated assistant") and is on by default. For customers whose WhatsApp number has an EU or EEA country code, or a Swiss or Gibraltar one, the line is always sent, whatever the setting says.

Three honest limits. First, Facebook and Instagram don't show Rumidza a customer's phone number, so the always-on rule can only apply to WhatsApp numbers; if you have European customers on Messenger or Instagram, keep the line switched on. Second, Rumidza's terms make you responsible for keeping that first line on where other local rules apply, such as some US states. The one-line approach is designed around the "first interaction" and "clear and conspicuous" wording of the EU, California and Maine rules, but Utah's safe harbour asks for disclosure at the start and throughout, which a single opening line doesn't provide. In Utah the base rule is answering honestly when asked, which is what Rumidza's assistant is instructed to do. Third, Rumidza doesn't add a new line when a teammate hands a chat back to the assistant, so the teammate should say so themselves, as in the example above.

If you want an assistant that answers instantly and is upfront about what it is, you can try Rumidza.

Sources

  1. EU AI Act Service Desk (European Commission) — Article 50: Transparency obligations for providers and deployers of certain AI systems
  2. EU AI Act Service Desk (European Commission) — Article 99: Penalties (official Italian-language version)
  3. EU AI Act Service Desk (European Commission) — Article 2: Scope
  4. European Commission — Guidelines on the transparency obligations for providers and deployers of AI systems (20 July 2026)
  5. Future of Privacy Forum — The AI Act implementation timeline: what changes under the AI Omnibus?
  6. Garrigues — AI Digital Omnibus Regulation has been published, redefining deadlines
  7. Addleshaw Goddard — EU AI Act: guidelines on transparency obligations (22 July 2026)
  8. California Business and Professions Code § 17940 (definitions)
  9. California Business and Professions Code § 17941 (bot disclosure)
  10. California Business and Professions Code § 17942
  11. California Business and Professions Code § 17943 (operative date)
  12. Justia — California Business and Professions Code § 17941 (2019), history note
  13. Utah Legislature — S.B. 226 (2025), Artificial Intelligence Consumer Protection Amendments, enrolled text
  14. Davis Polk — Utah scales back reach of generative AI consumer protection law (4 April 2025)
  15. Maine Legislature — Title 10, Required disclosure of use of artificial intelligence chatbot to engage in trade and commerce
  16. Verrill — Maine law now requires limited disclosures of artificial intelligence technology (30 June 2025)
  17. Meta for Developers — Messenger Platform and IG Messaging API policy
  18. WhatsApp Business Messaging Policy (last updated 23 September 2026)

Keep reading