wibbu

EU AI Act chatbot disclosure: a small business guide

Since August 2, 2026, AI that talks to customers in the EU must say it’s AI. Here’s what the rule asks for, what to say and a checklist.

Branko Conjic

In short

Since August 2, 2026, the EU AI Act requires that people know when they’re talking to AI. If AI replies to your customers in the EU, say so in the first message, clearly, in the chat itself. A note in your terms doesn’t count. For most businesses, it takes one sentence to fix.

The EU’s AI rules sound like something for big tech companies, and most of them are. But one rule reaches any business whose AI talks to customers in the EU. That includes a small shop with a website chatbot or AI replies on WhatsApp. The good news is that complying is simple.

Does it apply to you?

Ask yourself two questions:

  1. Does AI reply to customers on its own? For example, a website chatbot or an AI agent on WhatsApp or Instagram. Emails count too, if AI sends them without anyone checking.
  2. Can people in the EU reach it? Customers, website visitors or anyone messaging you from an EU country.

If both answers are yes, the rule applies to you, even if your business is based in the US, the UK or anywhere else. The Act covers AI whose output is used in the EU, wherever the company behind it is based.

If AI only drafts replies and a person sends them, see what happens with drafts below.

What the rule asks for

The rule is Article 50 of the EU AI Act. In plain terms:

  • When: people must be told they’re talking to AI at the latest when the conversation starts.
  • Where: in the conversation itself, in a “clear and distinguishable” way. A line in your terms of service, or a vague label like “assistant”, doesn’t count.
  • The exception: you don’t have to say it if it’s already obvious. Don’t rely on that, since saying it costs you nothing.
  • Who’s responsible: the main duty sits with the provider, the company behind the AI tool. But your customers see your business, not your vendor, so check that your tool says it, and switch the setting on if it’s optional.

What to say, channel by channel

Use the word “AI” and offer a way to reach a person. Here’s a first line for each channel:

ChannelFirst line
Website chat“Hi! I’m [business]‘s AI assistant. I can help with orders, hours and prices, or get you to someone on the team.”
WhatsApp“Hi, this is [business]‘s AI assistant. Ask me about bookings or opening hours. Type ‘person’ to reach our team.”
Instagram“Thanks for your message! I’m an AI assistant for [business]. I can answer quick questions, and our team can step in anytime.”
Email sent by AIA line at the top: “This reply was written by [business]‘s AI assistant. Reply ‘person’ to reach our team.”

The line should be clear in both directions. When a person takes over, have them introduce themselves: “Hi, this is Ana from the team. I’ll take it from here.” Customers always know who they’re talking to, and the switch feels like good service, not a trick.

What if AI drafts and a person sends?

Say AI writes a draft, and a person reads it, edits it if needed and sends it. Then the customer is talking to a person who used a writing tool. The chatbot rule is about AI that talks to people directly, so this setup generally falls outside it. Even the Act’s rule for AI-written news and public-interest texts doesn’t apply once a person has reviewed the text and takes responsibility for it.

Two caveats:

  • The review has to be real. If someone clicks “send” on every draft without reading it, that’s AI replying with extra steps.
  • This is our reading, not legal advice. If your business depends on getting this exactly right, check with a lawyer.

US chatbot disclosure laws

The US has no single federal rule, but a few states have their own:

  • California has had a bot disclosure law since 2019. It covers bots that hide what they are to push a sale or influence a vote. A clear disclosure keeps you on the right side of it.
  • Maine passed a chatbot law in 2025. Businesses can’t let an AI chatbot pass as a person. If customers could be misled, the chatbot has to clearly say it’s AI.
  • Utah requires businesses to say they’re using AI if a customer asks. Licensed professionals, like doctors or therapists, must say so up front when a chat involves health or financial details, or gives advice.

These rules keep changing. Saying it’s AI everywhere, from the first message, is the simplest way to stay ahead of all of them. For what AI should handle on its own and when it should hand over, see our guide to which customer messages AI should answer.

Your checklist

You can do all of this in an afternoon:

  1. List every place AI talks to customers: website chat, WhatsApp, Instagram, email and phone.
  2. Check that the first message in each one says it’s AI, using the word “AI”.
  3. Turn on your tool’s disclosure setting, or edit the greeting yourself.
  4. Make sure customers can reach a person, and say how in that first message.
  5. When a person takes over, have them introduce themselves by name.

Common questions

What are the fines? Up to €15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium-sized businesses, including startups, it’s whichever is lower. Adding a line to your greeting costs a lot less.

Does this apply if my business isn’t in the EU? Yes, if people in the EU can talk to your AI.

Do I need to label every AI message? The rule asks for a clear notice at the start of the conversation. Labeling AI messages is still a good idea, especially when AI and your team take turns.

Is this legal advice? No. It’s a plain-English overview to help you get started. For your situation, talk to a lawyer.

Keep reading