Does Your Chatbot Have to Say It's AI? The New Transparency Rules, Explained

Q: Who does this apply to?

The AI Act draws a line between providers (the companies that build AI systems, such as OpenAI or your chat vendor) and users (businesses that put someone else's AI to use, called "deployers" in the law). A webshop running an AI chatbot from Shopify, Intercom or Tidio is a user, and transparency duties apply to users too, the moment their AI speaks to the public or produces content for it. One key difference from the European Accessibility Act: Article 50 has no exemption for small businesses. Where the EAA exempts micro-enterprises (fewer than 10 staff and under 2 million euros in turnover) for services, the AI transparency obligation applies no matter how big or small your company is.

Q: What exactly is required, and by when?

Already in force (since 2 February 2025): AI literacy. Article 4 asks that staff who work with AI understand enough about it. No exam, but demonstrable basic knowledge that fits how you use AI. From 2 August 2026: the chatbot notice. A visitor has to know they are talking to AI no later than the first interaction. A clear notice in or above the chat window is enough; buried in the terms and conditions it is not. From 2 August 2026: marking AI content. AI-generated or AI-manipulated content has to be marked in a machine-readable way. For generative systems that were already on the market before that date, a grace period runs until 2 December 2026. Deepfakes and AI-written text aimed at the public have to be visibly labelled. Breaching these transparency obligations carries fines of up to 15 million euros or 3% of worldwide annual turnover, enforced by each national supervisory authority.

Q: What can you take care of now?

1. Take stock of your AI. Which chat widget is running, and is it AI-driven? Do you use AI for product copy, translations or images? Write down what you use, from which vendor and for what. That overview is your evidence later on. 2. Check your chat vendor. Ask whether the widget shows an AI notice at the first interaction, and whether it is switched on. With many vendors this is a setting, not a rebuild. 3. Mark your AI content. Ask your tooling about machine-readable marking (C2PA metadata, for example) and label the AI-written text you show to the public. 4. Give your team a little training. A short, documented explanation of what your AI can and cannot do covers most of the literacy obligation for a small organisation. Rather not work this out yourself? This is exactly what Seviranta's AI transparency check does: it continuously tests whether your chatbot notice and content marking are right, with the concrete fix included, for 29 euros a month (excl. VAT), on your plan or on its own. See the AI transparency check.

July 3, 2026

Short answer: yes, from 2 August 2026. That is the date EU AI Act Article 50 takes effect. If you run a chatbot or AI assistant on your website, you have to let visitors know they are talking to AI no later than the first interaction. And AI-generated content (text, images, audio) has to be marked as such in a machine-readable way. This touches almost every webshop that uses a chat widget or AI-generated copy.

Who does this apply to?

The AI Act draws a line between providers (the companies that build AI systems, such as OpenAI or your chat vendor) and users (businesses that put someone else's AI to use, called "deployers" in the law). A webshop running an AI chatbot from Shopify, Intercom or Tidio is a user, and transparency duties apply to users too, the moment their AI speaks to the public or produces content for it.

One key difference from the European Accessibility Act: Article 50 has no exemption for small businesses. Where the EAA exempts micro-enterprises (fewer than 10 staff and under 2 million euros in turnover) for services, the AI transparency obligation applies no matter how big or small your company is.

What exactly is required, and by when?

Already in force (since 2 February 2025): AI literacy. Article 4 asks that staff who work with AI understand enough about it. No exam, but demonstrable basic knowledge that fits how you use AI.
From 2 August 2026: the chatbot notice. A visitor has to know they are talking to AI no later than the first interaction. A clear notice in or above the chat window is enough; buried in the terms and conditions it is not.
From 2 August 2026: marking AI content. AI-generated or AI-manipulated content has to be marked in a machine-readable way. For generative systems that were already on the market before that date, a grace period runs until 2 December 2026.
Deepfakes and AI-written text aimed at the public have to be visibly labelled.

Breaching these transparency obligations carries fines of up to 15 million euros or 3% of worldwide annual turnover, enforced by each national supervisory authority.

What can you take care of now?

Take stock of your AI. Which chat widget is running, and is it AI-driven? Do you use AI for product copy, translations or images? Write down what you use, from which vendor and for what. That overview is your evidence later on.
Check your chat vendor. Ask whether the widget shows an AI notice at the first interaction, and whether it is switched on. With many vendors this is a setting, not a rebuild.
Mark your AI content. Ask your tooling about machine-readable marking (C2PA metadata, for example) and label the AI-written text you show to the public.
Give your team a little training. A short, documented explanation of what your AI can and cannot do covers most of the literacy obligation for a small organisation.

Rather not work this out yourself? This is exactly what Seviranta's AI transparency check does: it continuously tests whether your chatbot notice and content marking are right, with the concrete fix included, for 29 euros a month (excl. VAT), on your plan or on its own. See the AI transparency check.

The overlap with accessibility

There is one connection that often gets missed: a chatbot doesn't only have to say it is AI, it also has to be usable by everyone. A chat widget with no keyboard operation, or with no name for the screen reader, fails under the European Accessibility Act, exactly the kind of errors we run into in scans every day. If you are going to get your chatbot right, it is smart to handle both at once: the AI notice and the accessibility.

Where things stand right now

The European Commission is still working on guidelines and a code of practice for Article 50, so the finer points of how it plays out may still shift, and Brussels is discussing simplification packages that could touch the dates. This article reflects the state of things as of July 2026, with sources below. For legal advice on your specific situation, consult a lawyer.

Seviranta's AI transparency check does this for you: the same approach as our accessibility scan, but for the AI rules. Continuous testing, the concrete fix, an AI register as your evidence, and monitoring that spots new widgets before the supervisory authority does. For 29 euros a month, or 290 a year with two months free (excl. VAT), on your Seviranta plan or on its own without a plan. See and activate the AI transparency check. And don't forget: your chatbot already has to be accessible under the EAA today, and that is something we test right now. Scan your site for free.

Sources: Article 50, EU AI Act · AI Act implementation timeline · European Commission: guidelines and code of practice for transparent AI