AI chatbot disclosure is not fine print in your terms of service. It is a few sentences customers see before they type their first question. Those sentences need to make clear that software is answering, what that software will not do, and how to reach a person. When one of these pieces is missing, customers learn the truth only after a wrong answer, and by then their disappointment is aimed at your whole business.
This guide covers the first contact only: the greeting, the assistant name, the description of its limits and a visible route to your team. It does not explain how to hand an ongoing conversation to an agent or how to choose a tool. For the wider split of work between the model and people, read the guide on AI in customer service. Here you get four pieces of copy to adapt and a way to test them.
What EU rules say and when they apply
Article 50(1) of Regulation (EU) 2024/1689, the EU AI Act, requires AI systems intended to interact directly with people to be designed so that people are informed they are interacting with AI. The exception is when this is obvious to a reasonably well-informed, observant and circumspect person. Article 50(5) adds that the information must be clear, distinguishable and provided at the latest at the first interaction, in line with accessibility requirements. The provision formally addresses providers, yet the greeting on your website is usually written by you.
The wording of Article 50(1) and (5) comes from the regulation text on EUR-Lex. The article does not prescribe a ready-made sentence and does not settle how a specific widget should be assessed.
Article 50 applies from 2 August 2026. According to the European Commission, from that date the AI Office and national authorities in the Member States are responsible for supervising and enforcing the AI Act. This guide is not legal advice: if you are unsure whether your business acts as a provider or a deployer, ask a lawyer and your tool vendor. Rules in the UK, the US and elsewhere differ, so do not assume that this summary covers those markets or that EU-style copy satisfies their requirements.
The European Commission states that people using chatbots should be made aware they are interacting with a machine. The page does not assess the quality of specific messages and does not replace the regulation text.
Four places where customers learn about the AI
A single sentence in the greeting is easy to miss, especially on a phone or when someone returns to an old conversation. Spread the information across four elements, each answering a different customer question. Together they form a consistent picture: who is writing, under what name, within what scope, and what to do when that is not enough.
Scroll the table sideways to see all columns.
| Element | Customer question | Common mistake |
|---|---|---|
| Greeting | Am I talking to a person? | A generic "How can I help?" with no mention of AI |
| Assistant name | Who is on the other side? | A human first name and a stock photo |
| Knowledge limit | What can I expect? | A promise to answer every question |
| Human contact | What if this is not enough? | No information, or a promise nobody can keep |
Copy 1: a greeting that says it plainly
The greeting is the first and most important piece of information. The words "AI" or "automated assistant" belong in the first sentence, not three lines down after a sales pitch. Avoid euphemisms such as "smart helper" or "virtual consultant", because not every customer will read them as software. Keep it short and plain, and do not apologize for the technology.
"Hi, I am the store's AI assistant. I answer based on the information published on this website. If you would rather talk to someone from the team, just say so at any point."
Copy 2: an assistant name that does not pretend to be a person
The name shown in the chat header and next to every message reminds customers who they are writing to, even after they scroll through the history. A first name is acceptable when a clear "AI" sits next to it, but "Anna" alone with a photo of a smiling person suggests a human. Fictional example: a furniture workshop called its assistant "Anna from support", and customers asked it to call them at a specific time because they thought they were writing to an employee.
"Workshop AI assistant", "Woody, AI assistant" or "AI help". Use an icon or illustration rather than a portrait of a person. Keep the same name in the greeting, the header and the message label.
Copy 3: what the assistant does not know and will not do
A customer who knows they are talking to AI still does not know what to expect from it. State where the assistant gets its information and what it will not do: it will not change an order, grant a discount or decide on a complaint. That sentence prevents customers from treating a general answer as a binding decision. If the assistant often has no answer, look for the cause in the guide on why a chatbot cannot find an answer. On a clinic website that sentence should also rule out questions about symptoms, medication and test results, as the guide on AI chat for clinics explains.
"I know what is published on our website and in our help center. I cannot see your order and I do not make decisions about returns or discounts. Our team handles those requests."
Copy 4: a route to a person, visible from the start
An AI label loses its point if customers have nowhere to go when the answer does not help. At the start, one sentence saying they can ask for a person is enough, along with an honest note on when the team replies. Do not promise an instant connection if nobody works in the evening. The takeover itself, consent and the note for the agent are covered in the guide to the chatbot-to-human handoff.
"Want to talk to a person? Type that or use the button below. Our team replies on working days during the hours listed on our contact page."
By default, Meet AI only drafts suggestions for your team. In self-service or hybrid mode, a conversation with the assistant starts only when visitors choose the AI assistant card in the widget themselves. The default greeting includes the word "AI", and if you write your own text, the widget still marks the assistant and its messages with an AI badge. You set the assistant name yourself. When knowledge is missing, the assistant says so and offers a conversation with a person, and the handoff happens only after the customer agrees. Meet AI is available on the Scale plan.
Check the copy through a new customer's eyes
Before publishing, ask someone outside your team to open the chat on a phone and answer one question: who are they talking to? If they hesitate, rewrite the greeting rather than adding more paragraphs. You will find a full set of pre-launch checks in the guide to testing a chatbot before launch.
- The word "AI" appears in the first sentence of the greeting.
- The assistant name does not imitate a person and is identical everywhere.
- Customers know what the assistant will not do and where its information comes from.
- Asking for a person is explained and works on mobile too.
- The team hours shown in the chat match the real ones.
Honest AI labeling is part of the wider credibility of your website, alongside a visible address and realistic contact hours. The guide on building trust on your website shows how to find where customers lose confidence. Return to this checklist after every change to the greeting, the name or your opening hours.
Questions about chatbot disclosure
Is a line in the privacy policy enough? No. Article 50(5) of the AI Act calls for clear information provided at the latest at the first interaction, and customers rarely read a legal document before typing a question. Your policy can describe data processing in detail, but the chat itself needs a short, visible sentence. The simplest place for it is the greeting and the assistant name.
Will an AI label put customers off? We have no data that settles this, and figures without a source are not worth trusting. What is certain is that a customer who discovers the automation after the fact has a reason to feel misled. A clear greeting with a visible route to a person gives them a choice from the start. If you are worried about drop-off, watch how often customers ask for a person and which questions trigger it.
