Say it’s an AI.
Before they ask.
Yes: if a chatbot on your website is an AI, tell visitors so in the chat itself, from the first message. In the EU that has been a legal duty for chatbot providers since 2 August 2026. In Maine and New Jersey it falls on the business using the bot, and in Utah you must answer honestly when a customer asks. Here is who has to say it, what counts as saying it clearly, the lines to copy, and how the Builderify AI chatbot does it.
By the Builderify team Not legal advice
Where a rule exists, yes. The EU, the UK, Maine, New Jersey, Utah and California each get there by a different route, and they ask for slightly different things. What they share is the thing to avoid: a visitor who believes they are talking to a person when they are not.
So the practical answer is the same everywhere: say it plainly in the chat, say it at the start, and answer honestly if anyone asks. The rest of this page is the detail behind those three.
In the chat, in plain words, with or before the first reply. The EU asks for it at the latest at the first interaction; New Jersey and Utah’s safe harbor say “at the outset”.
A chatbot asked whether it is a person must not say yes. Utah makes it a duty, and the EU guidelines expect disclosure in every situation where the system is asked about its nature.
Not a duty for most small businesses, but the quickest way to settle a visitor who wanted a human. From 1 January 2027, California requires businesses with more than $500 million in annual revenue to offer a simple way to ask for one.
“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.”
It depends on the rule. The EU puts the duty on the company that makes the chatbot. Maine, New Jersey and Utah put it on whoever uses the bot with customers, which is you. The UK holds the business to account even when someone else supplies the tool.
Article 50(1) of the AI Act is addressed to providers: whoever develops a chatbot and offers it under their own name. If you use a vendor’s chatbot as it comes, that is usually the vendor. If you build your own, or rework someone else’s and run it under your name, it may be you.
Maine, New Jersey and Utah write their duties for whoever uses a bot with customers. Your vendor may supply the label, but the law points at the business that has the conversation, so the check is yours.
The CMA’s guidance says you are responsible for what an AI agent does in the same way you are responsible for what an employee does, even if someone else designed it or provides it on your behalf.
Whichever applies, the visitor never sees the provider. They see your chat. That is where the check belongs, and it takes a minute to do.
Most of these rules are written around where your visitor is, not where you are. Whether a particular rule reaches your business is a question for a lawyer.
These are the rules that most directly reach a customer-facing website chatbot, checked on 28 October 2026. They change often, so follow the links to the current text.
A bot must say so at the outset when selling or advertising.
An honest answer when asked, and a safe harbor.
A chatbot must not pass as human without notice.
Article 50(1) applies to chatbot providers.
Customer-service chatbots of large businesses.
The UK’s consumer rules have applied since 6 April 2025 and are not chatbot-specific, so they are in the table but not on the line. California’s older bot law dates from 2019.
The chatbot’s provider: whoever develops it and offers it under their own name. A provider outside the EU is covered when its output is used in the EU.
Design the chatbot so people are told they are dealing with an AI system, clearly and distinguishably, at the latest at their first interaction. Only waived when it is obvious to a reasonably well-informed, observant and circumspect person.
You. The CMA says a business is responsible for what an AI agent does in the same way as for an employee, even if someone else provides it.
Do not leave customers thinking a person is serving them. If the fact they are dealing with AI rather than a person might affect their decisions, tell them.
Anyone using a chatbot to deal with a consumer in trade or commerce. The law’s definition of a chatbot includes a web interface that simulates human conversation.
If the chatbot may mislead a reasonable consumer into believing they are dealing with a human, notify them in a clear and conspicuous manner that they are not. A violation counts as a violation of the Maine Unfair Trade Practices Act.
Anyone using an online bot to talk to people in New Jersey in connection with selling or advertising merchandise or real estate.
Say at the outset, clearly and conspicuously, that the conversation is with a bot. The Act defines a bot as an “automated online account”, so ask a lawyer whether a website chat widget counts.
A supplier using generative AI to deal with a consumer.
Say it is AI, not a human, when someone clearly and unambiguously asks. A safe harbor applies if the chatbot clearly and conspicuously says so at the outset and throughout.
§ 17941: anyone using a bot to mislead people about its identity to drive a sale. AB 1609: only businesses with more than $500 million in annual revenue.
Disclose that it is a bot, clearly and conspicuously, and § 17941 does not bite. From 2027, AB 1609 also bars large businesses from presenting a customer-service chatbot as human, asks for a clear in-chat disclosure when a reasonable person could be misled, and requires a simple way to ask for a human agent.
Any business.
“When using AI tools to interact with customers (think chatbots), be careful not to mislead consumers about the nature of the interaction.” This is guidance, not a chatbot law: the FTC Act’s ban on unfair or deceptive practices is what stands behind it. The post is no longer on ftc.gov, so the link goes to an archived copy.
Of worldwide annual turnover, whichever is higher. For small and medium-sized businesses, whichever is lower.
Of worldwide turnover is the most the CMA says you could be fined for breaking consumer protection law, and you could be forced to compensate affected consumers.
For a first offense, $5,000 for a second and $10,000 for each after. The Attorney General can also seek an injunction.
No enforcement action if the chatbot clearly and conspicuously says it is AI at the outset and throughout.
A wave of state laws since 2025 targets companion chatbots, built for ongoing, human-like conversations. California’s SB 243, Colorado’s HB 26-1263 and Iowa’s SF 2417 each leave out bots used for customer service or support, so a website support chatbot usually sits outside them. Oregon and Washington passed their own in March 2026, taking effect on 1 January 2027. Definitions differ, so read your own state’s before you assume.
The 2024 Colorado AI Act created consumer protections in interactions with AI systems. SB 26-189, signed on 14 May 2026, repeals and reenacts those provisions as rules for automated decision-making technology used in consequential decisions, such as employment, housing, lending and insurance, for decisions made from 1 January 2027. Pages that still cite the 2024 Act for a website chatbot are out of date.
The EU’s AI Omnibus gave generative AI systems already on the market until 2 December 2026, but only for marking their outputs under Article 50(2). The Commission’s guidelines say that disclosure by AI systems that interact directly with people must be ensured from 2 August 2026 (paragraph 153).
NCSL counted 1,539 AI bills introduced in the states in 2026 and 280 enacted by early October. An NCSL fellow told legislators that bills about letting users know they are engaging with AI had increased nearly threefold this year. NCSL’s AI legislation database is the place to check your own state.
This is general information, not legal advice. It applies each rule’s own words in plain language and nothing more. Whether a law reaches your business depends on where your visitors are and what your chatbot does. For a decision that costs money, ask a lawyer who works on AI and consumer law.
Plain words inside the chat, at the start. The Commission’s guidelines on the EU rule ask for information that is clear and distinguishable, at the latest at the first interaction, and give two examples for chatbots: one that starts the conversation by saying it is based on AI, and one that shows a label while also giving a first-turn disclosure.
The guidelines are non-binding, and only the EU’s Court of Justice can give the final reading. They are still the clearest picture there is of what regulators will look for.
The left list follows paragraphs 36 and 37 of the guidelines, the right list their paragraph 38. A name such as “Sarah from support” is our example of the “human-like representations that may mislead” they warn about.
The EU rule is waived when it is obvious to a reasonably well-informed, observant and circumspect person. The Commission says to interpret that exception restrictively, and its list of examples where it does not apply includes AI chatbots embedded in online platforms or helpdesk tools whose replies users may perceive as human-generated. A support chatbot on a business website is that case, and a profile picture that shows a human makes the interaction less obvious still.
The Commission says a single, prominent notice before the first interaction is likely to suffice in most instances. In riskier settings, such as financial, insurance, legal or health advice and complaints handling, or when vulnerable people may be talking to the chatbot, it expects periodic reminders too. It also asks for the notice to be accessible to people with disabilities, which is one more reason to write it as text in the chat, not bake it into an image. For the wider rules, see the European Accessibility Act guide.
Both points are from the Commission’s Article 50 guidelines (paragraphs 34, 40 and 45).
Tell it about your business and it writes a first message, a label, an answer to “are you a real person?” and a line for your policy or footer, in English, German, French or Spanish. It runs in your browser; nothing you type is sent anywhere.
Live preview: four lines in English
The chat on this website is an AI assistant, not a person. It answers from the content of this website and can make mistakes.
The wording is a starting point, not legal advice. Change anything that is not true of your chatbot: if it cannot hand over to a person, choose “Point to a contact” rather than promise one. The German, French and Spanish lines are written to be read by a native speaker before they go live.
It says so by default, and you cannot switch that off. The AI Chatbot shows the label the moment a chat opens. What you add is the wording around it, in about ten minutes.
Open your chat the way a visitor would. The header shows your assistant’s name with “AI assistant” beneath it from the first second, and the name-and-email form, if you use one, says “You’ll be chatting with an AI assistant.” There is no setting to switch the label off.
In the chatbot’s settings, under Identity, set the Assistant name. A role such as “Fernhill Assistant” is clearer than a first name, which can read as a person. The name is shown to visitors and used in the prompt.
Under Guardrails, add the Privacy policy link. It appears under “AI assistant” in the chat, so visitors can see who handles their messages.
In your Bubble, open the AI Chatbot block and paste the first message from the builder above into Welcome Message. It shows when the chat opens, in the words your visitors read.
In Knowledge, add a Q&A for “Are you a real person?” with the answer from the builder. What the assistant says to that question depends on what it has been taught, so do not assume: add it, then test it below.
Under Guardrails, leave Offer to hand over to a human on. A “Talk to a person instead” link appears under the box once the chat has started, and the conversation goes to your team. The hand-over how-to follows it all the way.
People answering in the chat is Live Chat, and email follow-ups sit in Support Tickets. On WordPress? Here is how to add the chatbot.
The label is built in, but what the assistant says when someone asks depends on what it has been taught. Put each question to your own chat, in your own words and in your visitors’ language, and tick what passes.
Nothing here replaces advice for your own situation, and a label does not make a site compliant: your privacy policy, consent and other legal duties are separate.
Ask each one in your own chat, then tick what passes.
Every account gets 250 usage credits free each month, and the chatbot is billed at what each answer actually cost, with a running total on every conversation. There is no subscription: if you outgrow the free credits, one-time packs start at $9 and never expire. See the pricing page.
Start freeWhere a rule applies, yes. Since 2 August 2026 the EU's AI Act has required chatbot providers to make sure people are told they are dealing with an AI system, Maine and New Jersey require the business using the bot to say so, and Utah requires an honest answer when a customer clearly asks. Saying it plainly in the chat, from the first message, is the practice all of them point to. This is general information, not legal advice.
It depends on the rule. In the EU, Article 50(1) of the AI Act is addressed to the chatbot's provider, the company that offers it under its own name. In Maine, New Jersey and Utah the rules point at the business using the bot with customers, and in the UK the CMA says a business is responsible for an AI agent as for an employee, even when someone else provides it. Check what your chatbot shows, whoever's job it is.
Not on its own. The European Commission's guidelines on Article 50 say a disclosure that sits only in terms and conditions, URLs or documentation is insufficient, though it can complement one in the chat. They look for clear information at the latest at the first interaction, so put it where the conversation happens.
Don't count on it. The EU rule has an exception for interactions that are obvious to a reasonably well-informed, observant and circumspect person, but the Commission says to interpret it restrictively, and its examples of cases where it does not apply include AI chatbots in online platforms or helpdesk tools whose replies users may perceive as human-generated. A support chatbot on a business website is that case.
Yes. The chat header always shows "AI assistant" under your assistant's name, with an optional link to your privacy policy, and there is no setting to remove it. You write the welcome message, can add an "Are you a real person?" question and answer for the assistant to use, and a "Talk to a person instead" link appears under the box once the chat has started, while hand-over is on, which it is by default. The built-in label is in English, so say it in your visitors' language in the welcome message too.
Legal facts checked on 28 October 2026. Laws change, so read the current text before you rely on a date or a figure here.
More leads from the traffic you already have, and follow-up that sends itself. Free to start. No card, no monthly fee.