GuideGuides and tutorials
Must You Tell Customers They're Chatting With AI?
In the EU the answer has been yes since August. In the US it depends on the state. The laws in one table, and the wording that keeps the conversation friendly on chat, WhatsApp and the phone.
The short answer
Often, yes. If people in the EU chat with your AI, Article 50 of the AI Act has required since 2 August 2026 that they be told they are talking to an AI, at the latest in the first interaction. The US has no federal rule, but Maine requires disclosure whenever a customer could mistake the bot for a person, Utah when a customer asks, and California bans hiding a bot to push a sale. The safe habit everywhere: say “AI” in the first line.
A customer asks your website chat about a late parcel, gets a quick, specific answer, then types: “Am I talking to a real person?” What the chat says next is now a legal question in a growing list of places, not only a matter of manners.
Below: the rules for the places most small businesses sell to, each linked to its official text, then wording you can paste into your chat, WhatsApp and phone greetings, and a seven-point checklist you can finish this afternoon.
General information, not legal advice. Laws were read on their official sources on 20 October 2026, and several are new or not yet in force. If your business is regulated (health, legal, financial) or you are unsure how a rule applies to you, ask a lawyer where your customers are.
The laws, place by place
What each rule asks of a business running an AI chat or voice assistant for customers. Read the “who it covers” column first: several new laws are about companion apps, not support.
| Where | Law | Who it covers | What you must say, and when | Penalty |
|---|---|---|---|---|
| European Union | AI Act, Article 50 | AI systems that talk directly with people in the EU, wherever the business is based | That it is an AI, clearly, by the first interaction at the latest, unless obvious. Since 2 August 2026 | Up to €15m or 3% of worldwide turnover; for small and medium firms, whichever is lower |
| California | BOT Act, Bus. & Prof. Code 17940 to 17943 | Anyone using a bot online to talk with people in California | Hiding a bot to push a sale or a vote is unlawful; a clear, conspicuous disclosure removes liability. Since 1 July 2019 | None in the chapter itself; other laws still apply |
| Utah | AI Policy Act, amended 2025 | Generative AI in consumer transactions; licensed professions | That it is AI when a customer clearly asks; up front for licensed professions in high-risk talks | Administrative fines up to $2,500 per violation |
| Maine | 10 M.R.S. §1500-DD (2025) | Any AI chatbot used to trade with consumers | That it is not a human, clearly and conspicuously, whenever a reasonable consumer could think otherwise | A violation of the Unfair Trade Practices Act |
| Washington | ESHB 2225 (2026) | Companion chatbots; customer service bots excluded | Not human, at the start and every three hours. From 1 January 2027 | Consumer Protection Act |
| Colorado | SB 26-189 (2026) | Automated tools in decisions on jobs, housing, lending, insurance, health care and more | Notice at the point of interaction with a covered tool. From 1 January 2027 | Attorney General, Consumer Protection Act |
| US, federal | FTC Act, Section 5 | Any business selling to US consumers | No AI rule; a practice likely to mislead a reasonable consumer on something material can be deceptive | FTC enforcement |
The EU: Article 50, in force since 2 August
Article 50(1) of the AI Act says AI systems meant to interact directly with people must be designed so those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed, observant person. Article 50(5) adds the when and how: clearly and distinguishably, at the latest at the first interaction, and accessibly.
The date held. The Digital Omnibus, Regulation (EU) 2026/1744, moved several later deadlines, mainly for high-risk systems, but Article 113 still applies Article 50 from 2 August 2026. Its only Article 50 transition, Article 111(4), gives generative systems already on the market until 2 December 2026 to add machine-readable marks to their output. The duty to say “I am an AI” was not moved.
Strictly, Article 50(1) is written for the provider, the party that develops a system or puts it into service under its own name (Article 3). In practice the business that switches the assistant on writes its name and first line, so the disclosure customers see is yours to get right. Article 2 reaches providers and deployers outside the EU when the output is used in the EU, and Article 99 sets the fines in the table. The Commission published guidelines on Article 50 on 20 July 2026. For risk tiers, GDPR and the wider duties, our EU AI Act guide for chatbots goes deeper.
The US: a patchwork, mostly aimed elsewhere
California’s BOT Act, operative since 1 July 2019, makes it unlawful to use a bot to talk with someone in California with the intent to mislead them about its artificial identity, to encourage a purchase or influence a vote. An operator that discloses is not liable, if the disclosure is “clear, conspicuous, and reasonably designed” to inform. The test turns on intent; the way out is simply to say it.
Maine has the plainest rule. 10 M.R.S. §1500-DD bars using an AI chatbot, defined as software that simulates conversation through text or speech, in a way that may lead a reasonable consumer to believe they are talking with a human, unless they are told clearly and conspicuously that they are not. A breach is an unfair trade practice, for which the Attorney General can seek an injunction and restitution (5 M.R.S. §209).
Utah narrowed its 2024 law in 2025. According to the Future of Privacy Forum’s April 2025 overview of the enrolled bills, a business using generative AI must say it is AI, not a human, when a consumer makes a “clear and unambiguous request” to know; licensed professions must disclose up front in “high-risk” conversations. Disclosing at the outset and throughout earns a safe harbor. Fines reach $2,500 per violation, and SB 332 moved the law’s repeal date to July 2027.
Colorado replaced its 2024 AI Act with SB 26-189, signed 14 May 2026, which from 1 January 2027 covers automated tools used in consequential decisions such as hiring, housing, lending and insurance. A chat assistant answering delivery questions sits a long way from that.
The 2026 wave is mostly about companion chatbots, apps built to keep up a relationship with a user. Orrick’s April 2026 roundup lists new laws in Oregon, Washington, Idaho, Nebraska and Tennessee. The two we read in full carve out support bots: California’s SB 243 excludes “a bot that is used only for customer service”, and Washington’s ESHB 2225 excludes customer service bots that do not sustain a relationship across conversations. Check each new law’s definitions before assuming you are in or out.
Federally, there is no AI disclosure rule, but the FTC’s Policy Statement on Deception (1983) still frames the question: a practice or omission likely to mislead a consumer acting reasonably, about something material, is deceptive. A bot that lets a customer believe a person checked their refund could fit that.
How to disclose without killing the conversation
Most worry about disclosure comes from picturing a legal notice: “You are interacting with an automated system.” A good disclosure does three jobs in one friendly sentence: it says AI, it says where the answers come from, and it says how to reach a person. The examples use Lark & Loom, a homeware shop we made up for this guide.
Website chat, the opening message
"Hi, I'm the AI assistant for Lark & Loom. I answer from our own pages, and a person can take over any time. What can I help with?"
Says AI, says where the answers come from, and says how to reach a person.
WhatsApp, the first reply
"Thanks for messaging Lark & Loom. This is our AI assistant replying. Ask for a person at any time and our team will pick this up."
The customer writes first, so the disclosure rides on the first answer, with the escalation route WhatsApp asks for.
Phone line, the first words
"Hello, you're through to Lark & Loom. I'm the shop's AI assistant, and I can help with orders, delivery and returns. To reach the team, you can also email hello@larkandloom.example."
Heard before the caller says a word, so nobody explains a problem to something they took for a person.
Where to put it
- In the first message, not the terms. A line in your privacy policy is not where a customer meets the bot.
- In the name on the window. “Lark & Loom AI assistant” stays visible all the way down the conversation. A friendly name is fine; keep “AI” beside it.
- In the answer to “are you human?” This is the moment Utah’s rule is about, and the moment a customer decides whether to trust you: “I’m an AI assistant. Would you like me to bring in a member of the team?”
Our guide to writing chatbot greetings has more openers by industry; each works with “AI” added.
On the phone line
A natural voice is easy to mistake for a person, and a caller cannot scroll back up. Say it in the first sentence, short enough to survive a noisy kitchen, then say what the assistant can help with so callers with something else can choose another route at once. Our comparison of AI voice agents and chatbots goes further.
On WhatsApp
WhatsApp has a rule of its own. Its Business Messaging Policy (last updated 23 September 2026) allows automation in the 24-hour customer service window only if “prompt, clear, and direct escalation paths” are available, such as a human agent in the chat, a phone number, email or a support form. Put the disclosure and the route to a person in the first reply.
Worked example: what is at stake for a small shop
Lark & Loom, our invented shop, is based in Porto, turns over €3 million a year and sells across the EU. Under Article 99 the ceiling for an Article 50 breach is the higher of €15 million or 3% of turnover, but for a small or medium firm it is the lower: 3% of €3 million, so €90,000. That is a legal maximum, not a forecast.
Against that, the shop has four lines to rewrite: the chat opening message, the window’s name, the WhatsApp first reply and one sentence in the assistant’s instructions about “are you a real person?”. At a few minutes each, plus a test message on each channel, that is well under an hour.
A 7-point AI disclosure checklist
List every place your AI talks. Website chat, WhatsApp, email, Telegram, the phone line. Each starts differently, so each needs its own first line.
Put the word "AI" in the first message on each. Not "virtual assistant", which has long meant remote human staff. The EU sets the first interaction as the latest moment.
Label the chat window. A name such as "Lark & Loom AI assistant" keeps the fact visible after the greeting scrolls away.
Make it readable and audible. Article 50(5) asks for accessible information: plain text a screen reader can read, not an icon alone.
Offer a person, and test it. Ask for a person on each channel yourself and time how long a reply takes.
Tell the AI never to claim it is human. Write it into its instructions: asked if it is a person, it says it is an AI assistant and offers the team.
Keep a dated record. Save each channel's greeting with the date you changed it, and recheck the laws twice a year. Several US rules start in 2027.
Disclosure is half of being straight with customers. The other half is what happens when they take up the offer of a person: our guide to chatbot hand-overs covers that, and the GDPR and HIPAA chatbot guide covers the personal data a conversation collects.
Where EzyConn fits
In EzyConn the business writes every line that does the disclosing, so the AI can say it is an AI from its first words:
- Website chat. You write the Opening message customers see before they type, the assistant’s Name (where “AI assistant” can sit) and its Reply to a hello. Its Instructions, in plain sentences, are where you tell it to answer “are you human?” honestly and offer the team.
- Phone line. From the Starter plan, What it says when it picks up is the first thing a caller hears on your own Twilio number. See the AI phone line.
- WhatsApp, email and Telegram. The same assistant answers from the same knowledge. Write the disclosure into its instructions, then message each channel from your own phone and read the first reply as a customer would. See the WhatsApp integration.
The assistant answers from your website, files and the exact answers you write. When it does not know, it says so and offers a person; when the customer accepts, it steps back and your team gets the whole conversation in the inbox or in Slack. The pre-chat form can include a Consent to store their details checkbox alongside name, email, phone and department, and you choose the fields.
The Free plan includes the AI agent, needs no card and does not expire, so you can put a free AI chatbot on your website, write your disclosure and ask it “am I talking to a person?” yourself. Plan limits are on the pricing page, or start free now.
AI chatbot disclosure: common questions
Do I legally have to tell customers they are talking to a chatbot?
If your chatbot talks to people in the EU, yes: since 2 August 2026, Article 50 of the AI Act requires them to be told they are interacting with an AI, at the latest at the first interaction, unless it is obvious. In the US it depends on the state. Maine requires it whenever a customer could mistake the bot for a person, Utah when a customer clearly asks.
Did the Digital Omnibus delay the EU chatbot disclosure rule?
No. The Digital Omnibus, Regulation (EU) 2026/1744, pushed back several later deadlines, mainly for high-risk systems, but Article 113 still applies Article 50 from 2 August 2026. Its one Article 50 transition covers machine-readable marking of AI-generated content: systems already on the market before 2 August 2026 have until 2 December 2026 for that.
Does the EU AI Act apply to a business outside the EU?
It can. Article 2 covers providers that place AI systems on the EU market wherever they are based, and providers and deployers outside the EU where the system's output is used in the EU. If people in the EU chat with your assistant, plan as if the duty applies.
Do I have to disclose AI on a phone line?
The EU rule covers any AI system that interacts directly with people, and Maine defines a chatbot as software that talks through text or speech. Treat a voice agent like a chat window: the caller should hear that it is an AI in the first sentence, before they start explaining their problem.
Do I have to offer a human as well?
Article 50 asks for disclosure, not a hand-over. WhatsApp does ask for one: its Business Messaging Policy allows automation in the 24-hour window only with prompt, clear and direct escalation paths, such as a transfer to a human agent, a phone number, email or a support form.
What is the fine for not disclosing an AI chatbot?
In the EU, up to €15 million or 3% of worldwide annual turnover, whichever is higher, with small and medium firms capped at whichever is lower. Utah allows administrative fines of up to $2,500 per violation. Maine treats a breach as an unfair trade practice. These are legal ceilings, not typical fines.
Do I need to repeat the disclosure during the conversation?
The EU text sets the latest moment, the first interaction, rather than a repeat. Utah gives a safe harbor to businesses whose AI discloses at the start and throughout. A first line that says AI, plus a label that stays on the chat window such as "AI assistant" next to its name, covers both.
Keep reading
- The EU AI Act and chatbots: risk tiers, GDPR and deployer duties
- How to write chatbot greetings people answer
- How accurate are AI chatbots, and how to check yours
- Building customer trust in AI support
- How EzyConn works, from your content to the hand-over
Last checked against the official texts linked above. This article is general information, not legal advice.