Open your website on your phone and type "hi" into the chat. Look at what sits directly above the text box.
On most UK small-business sites it is one of three things: nothing, a first name with a stock photo, or a pale grey line reading "Virtual assistant — how can I help?".
What people actually type into Google is whether any of that is against the law. The honest answer has three parts, and only one of them mentions AI at all.
No UK statute makes you say it
There is no UK AI Act. The approach has been to leave AI to the regulators that already exist — the ICO on data, Ofcom on online services, the FCA in financial services, the CMA on consumer protection — rather than pass one law covering it. A private member's bill, the Artificial Intelligence (Regulation) Bill [HL], has been introduced in the Lords more than once and has never become law.
So no line in a British statute says "tell people they are talking to a machine". Most articles stop there, and that is where they mislead you. Three other things fill the gap.
The EU rule that reaches UK businesses — from 2 August 2026
Article 50 of the EU AI Act, Regulation (EU) 2024/1689, requires that people be told they are interacting with an AI system. It applies from 2 August 2026.
It is EU law and you are not in the EU. That does not settle it. Article 2(1)(c) extends the Regulation to providers and deployers established in a third country "where the output produced by the AI system is used in the Union", and the Commission's guidelines of 20 July 2026 apply that to third-country deployers where use of the output in the Union is foreseen, directed or authorised by them.
In plain terms: if your chat handles enquiries from customers in Ireland, Spain or anywhere else in the bloc, this is pointed at you. Plenty of UK trades and agencies are in that position without having thought about it.
Worth knowing what did not happen. The Digital Omnibus regulation — Regulation (EU) 2026/1744 of 8 July 2026 — pushed the high-risk obligations back to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I. Article 50 kept its original date. The one slippage that touches it is narrow: systems already on the market before 2 August 2026 have until 2 December 2026 to mark synthetic content in machine-readable form under Article 50(2). The "this is an AI" notice itself gets no grace period.
The ceiling for getting it wrong sits in Article 99(4)(g): up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups, Article 99(6) flips that to whichever is lower.
What UK law already asks, without using the word AI
Transparency under the UK GDPR. Articles 5(1)(a), 13 and 14 require people to know who is processing their data and why. A bot presented as a named member of staff is hard to square with that, and the answer has to be visible in the interface, not buried in a privacy notice nobody opens.
Misleading practices. The unfair commercial practices provisions of the Digital Markets, Competition and Consumers Act 2024 came into force on 6 April 2025. They let the CMA fine a business directly — up to 10% of worldwide turnover — without going to court first. A misleading action includes creating a false impression that causes the average consumer to take a decision they would not otherwise have taken. Presenting an AI as a person is the shape of thing that regime was written around.
Be careful with anyone who tells you more than that. We could not find a published CMA case about a chatbot posing as a human. The exposure is real but untested, and anyone quoting you a UK fine figure for it is guessing.
What "clear" actually looks like
This is where nearly all the coverage evaporates: lots of risk-classification tables, no sentence you can copy. The Commission's July 2026 guidelines are blunt about what fails — references buried in terms and conditions, a mention in the website footer, or a vague label are unlikely to be enough. Article 50(5) adds the timing and the form: clear and distinguishable, at the latest at the first interaction, and meeting applicable accessibility requirements.
Web chat. A fixed line above the text box, visible without scrolling, at readable contrast. Something like: "AI assistant for [company]. You're talking to an AI, not a person. Say the word and I'll pass you to the team."
WhatsApp. There is no interface to decorate, so the notice goes in the first message you send. The first one, not the second.
Voice. If the assistant speaks, the notice is spoken. The guidelines cover voice assistants explicitly.
The direct question. "Are you a real person?" has one correct answer. Make that a system rule rather than a line in the prompt, because a model will drift away from a suggestion.
What you do not need is a legal paragraph. One honest sentence performs better and complies better.
Who is on the hook — you or your vendor
A distinction most write-ups get loose. Article 50(1) binds the provider of the system, not the deployer; the deployer duties in Article 50 are the other ones, covering deepfakes, emotion recognition and public-interest text.
But "provider" under Article 3(3) means whoever develops a system, or has one developed, and places it on the market or puts it into service under their own name or trade mark. Take a third-party assistant, put it on your site as "[Your company] Assistant", and the line gets blurry fast.
Two practical consequences. Get it in the contract that the disclosure ships as standard and is editable. And assume that in front of your customer — and in front of the CMA — the one explaining it is you.
What none of this fixes
- It does not require consent. It is information, not permission.
- It does not make your chat a high-risk system. That is a separate category, and it is the one that just moved to 2027 and 2028.
- There is no shortcut via the code of practice. The Commission confirmed the Code of Practice on Transparency of AI-Generated Content in July 2026, but it covers the content-marking duties in Articles 50(2), (4) and (5) — not the chatbot disclosure in 50(1). For chatbots there is nothing to sign that buys you a presumption of compliance.
- It says nothing about whether your chat is right. You can satisfy Article 50 to the letter and still run an assistant that invents a price. That gets punished by the customer, not by Brussels.
- It does not replace the UK GDPR, which asks its own questions separately.
Five minutes, three checks
- Open your chat on a phone, logged out. Is the notice visible before you type, without scrolling?
- Message your business WhatsApp from a different number. Is the disclosure in the first reply?
- Ask it "are you a real person?". If it hedges, deflects or says yes, you have a bigger problem than Article 50.
Zatio ships this as standard: the notice appears before the first message on both web chat and WhatsApp, and the assistant admitting it is an AI is a system rule rather than an instruction to the model. What we do not do is your compliance — we do not write your privacy notice, we do not pick your lawful basis, and any other channel you have open is yours to sort. If you want to see it working, here is the product.
This is general information, not legal advice. For compliance decisions, talk to someone who can look at your specific case.