From 2 August, your chatbot must identify itself — what this means for your site

On 2 August 2026, a new part of the European AI Act takes effect: anyone running a chatbot or AI assistant on their website must explicitly inform visitors about it. Not a checkbox buried in the footer, but a clear notice at the moment someone starts interacting with the system.

What the law actually requires

The transparency obligation in Article 50 of the AI Act applies to any system that communicates directly with people where it isn't immediately obvious that AI is involved. A chatbot answering questions about delivery times, an AI assistant helping draft quotes, a system that auto-replies to reviews: if a visitor could reasonably think they're talking to a human, that has to be corrected. A simple line at the top of the conversation, a label next to the chat name, or a short introductory sentence from the bot itself is enough.

Why this is more than a legal footnote

For many small business owners, AI still feels like something bolted on: a chat plugin, an automated email reply, a paragraph a language model generated. The law now forces a different question: where in the business does a visitor actually talk to AI, and is that clear everywhere? Many businesses simply don't have that overview, because AI features were added one at a time — a tool here, a plugin there — without anyone mapping the whole picture.

What this means in practice

For a website chatbot, the fix is usually small: an opening line like "You're chatting with our AI assistant" is enough. It gets trickier with less visible systems — AI that auto-replies to emails on the business's behalf, or generated content presented as editorial writing. There the question isn't just "how do we disclose this" but "where are we actually using AI, and does that use still fit how we want the business to come across".

The opportunity behind the obligation

Businesses that use AI transparently and deliberately don't need to see this law as a burden. Customers generally respond fine to a chatbot, as long as it's clear what to expect from it. A clear notice also prevents a common source of frustration: someone who thinks they're chatting with a staff member and gets a generic answer is more annoyed than someone who knew upfront they were talking to an assistant and appreciated the speed.

This is a good moment to zoom out: which AI features are actually live, who added them, and do they still work together or have they become disconnected islands. More on that approach under AI & automation.