In October, the owner of an online store selling dietary supplements receives a message from her marketer: „We need to adjust the chatbot; otherwise, we risk a fine.“ But no one told her what kind of fine it was, when the rule went into effect, or whether it even applied to her.
She Googles „AI Act“ and finds dozens of legal articles full of references to sections of the law, but no one tells her what she should actually do with her chatbot on Shoptet.
This is not an isolated case. Starting August 2, 2026, new rules on artificial intelligence (AI) transparency will take effect in the European Union, affecting every website with a chatbot or content written or generated using AI. This isn’t a future threat—the rule has been in effect for over two months, and companies are only now catching up.
Let's break down exactly what you need to do, why those who claim the deadline has been postponed are wrong, and, most importantly, how to technically adjust the chatbot on your website so that it works properly.
What exactly does Article 50 of the AI Act require of you?
Article 50 of the AI Act—that is, European Union Regulation 2024/1689—requires one main thing of you: if your website has a chatbot or another system that communicates directly with visitors, it must be clear to the user that they are talking to a machine, not a human. This requirement takes effect on August 2, 2026.
There is an exception, but it is narrower than it seems. You are not required to provide this information only when it is obvious to a reasonably informed and attentive person—that is, when the style and circumstances of the communication clearly indicate that the other party is not a human being. In practice, this applies, for example, to a simple pop-up window with pre-set buttons such as „I want a discount code“ or „Track my order.“ However, as soon as a chatbot engages in a free-flowing conversation and responds in its own words, the exception no longer applies, and you must provide the information clearly, namely no later than the first interaction.
The same article addresses three other situations that may indirectly affect you. Audio, image, and video content created by artificial intelligence must be technically labeled as AI-generated; this is primarily the responsibility of the tool’s manufacturer, not you as a website operator. As the publisher, you must always report deepfake content. And you must report AI-generated text that you publish to inform the public about matters of public interest if it has not undergone human editorial review.
| Situation | Do you have to report it? |
|---|---|
| Free-form Chatbot | Yes, during the first interaction |
| A simple menu with predefined buttons | No, if it's obvious that it's not a person |
| AI-generated text on a blog with human editorial oversight | No |
| Unedited AI text informing the public about a matter of public interest | Yes |
| Deepfake (a doctored photo or video of a person) | Yes, always |
In practice, then, two questions are decisive. Does the system engage in a free-flowing conversation where the user doesn’t know in advance what they can ask? And is it clear from the context that it’s not a human? A chatbot that responds to the question „When will my package arrive?“ in its own words based on a specific order is conducting a conversation. A simple search tool in the help section that merely filters through pre-prepared answers does not lead the conversation, and that is the crucial difference.
Did you think it had been postponed? Only partly.
On July 27, 2026, the so-called AI Omnibus—a package through which the European Commission simplified part of the AI Act—took effect. Many companies took away only one message from this: „The AI Act is being postponed.“ That’s not entirely true, and that difference could cost you dearly.
The AI Omnibus has indeed pushed back the deadlines, but only for the so-called high-risk systems listed in Annex III of the regulation—such as AI used in human resources, credit scoring, or critical infrastructure. The rules for these systems will not take effect until December 2, 2027, or rather August 2, 2028 for AI embedded in products such as elevators or machines. The transparency requirements under Article 50—specifically those pertaining to chatbots and AI-generated content—were not affected by the postponement. They remain in effect as of August 2, 2026, without change.
In other words, a typical e-commerce chatbot has never been—and is not—a high-risk system as defined by the regulation, so the postponement under the Omnibus Directive does not apply to it at all.
Does this apply to small businesses as well, without exception?
Here’s a distinction worth highlighting, because companies often confuse it with another law. The Web Accessibility Act includes an exception for micro-enterprises—that is, companies with up to 10 employees and an annual turnover of up to 2 million euros are not subject to this requirement. We wrote about this in detail in article on website accessibility.
Under the AI Act’s transparency requirements, there is no such “safety net” exception based on company size. The rule that a person must know they are speaking with an AI applies equally to a multinational e-commerce site and to a sole proprietor using a chatbot that costs 500 Kč per month.
The size of a company matters in only one respect: the amount of the fine. Small and medium-sized enterprises face the lower of two penalty caps, not an exemption from the obligation itself.
Who Enforces These Rules in the Czech Republic, and What Are the Penalties for Violations?
Starting August 2, 2026, oversight of the AI Act in the Czech Republic will be shared among three agencies. The Czech Telecommunications Office (ČTÚ) is the primary supervisory authority, the Czech National Bank (ČNB) oversees the financial sector, and the Office for Personal Data Protection (ÚOOÚ) oversees cases where AI systems involve personal data—and a chatbot that reads a customer’s name or email address almost always involves such data.
Penalties for violations of transparency obligations are set forth in Article 99(4) of the Regulation: up to 15,000,000 euros or 3 % of global revenue companies for the most recent fiscal year, whichever is higher. For small and medium-sized enterprises, the lower of these two caps applies.
Before you panic, here’s some reassuring information straight from the Czech Telecommunications Office (ČTÚ) website: the agency currently focuses primarily on education, not on handing out fines. Furthermore, as of the date of this article’s publication (October 23, 2026), the Czech Republic has not yet adopted a national implementing law that would regulate the specific procedure for imposing fines. However, the EU regulation applies directly, regardless of whether the Czech Republic has its own implementing law, so relying on the idea that „it doesn’t apply here yet“ is a fallacy.
Our Approach: How to Customize a Website Chatbot, Step by Step
This is the part that legal articles usually leave out, even though it's the most important part—how you actually get the information onto the website.
- Edit the chatbot's first message. Whether you use Tidio, Smartsupp, Dialogflow, or your own solution connected to a language model, the first sentence of the conversation should be clear, such as: „Hi, I’m the AI assistant for the [name] online store. I can help you with your order, shipping, or returns, and for more complex questions, I’ll connect you with a support representative.“ It’s not enough to just include this in the terms and conditions—it needs to be visible right in the conversation.
- Add a permanent visual label. In addition to the introductory message, add a short caption to the icon or chat header, such as „AI Assistant“ or „Virtual Assistant“. A user who joins the conversation later or misses the opening message must be able to tell who they're talking to, even without reading the chat history.
- Set up a clear handoff to the person. A chatbot should be able to recognize when a query is beyond its capabilities and offer to transfer the customer to a live agent or email support. This reduces customer frustration as well as legal risk.
- Please review the Terms of Use and Privacy Policy. If a chatbot stores a name, email address, or order number, it is processing personal data, which falls under the GDPR, the General Data Protection Regulation (we wrote about this in more detail in article about cookies on the website). Add a sentence to the terms and conditions stating that communication takes place with an AI system and explaining how, if necessary, the communication is transferred to a human.
- Try it out for yourself as a guest. Open the website in an incognito window and ask the chatbot a common question. Without any hints, you should realize within a few seconds that you're talking to a machine.
An interesting note about the Czech market: one of the most widely used chat tools on the websites of small and medium-sized businesses, Smartsupp, is a Czech company based in Brno, so the change will likely be aided by documentation and support written in Czech, not just general English instructions.
What to Do About AI-Generated Content on Your Blog and in Images
The second part of Article 50, which companies often overlook, does not concern chatbots, but rather content. If you publish text on a blog that was written or edited by artificial intelligence and undergoes human editorial review—meaning someone actually reads it, edits it, and takes responsibility for the content—the notification requirement does not apply to you. However, as soon as you publish AI-generated text without human editorial review and it serves to inform the public about matters of public interest, you must disclose that it is AI-generated.
The same logic applies to images. The primary technical identifier—that is, the watermark embedded directly in the image data—is typically handled by the tool’s developer, such as ChatGPT or Midjourney, not by you as the website operator. However, as the publisher, you are responsible for ensuring that the content you knowingly post on your website is not misrepresented as something other than what it is. An AI-generated illustrative image in the hero section of a blog is typically handled simply by including a caption or alt text that identifies it as illustrative.
Moreover, having a human editorial team is not just a legal safeguard. Content backed by a specific author with professional expertise is also at the heart of how E-E-A-T builds trust with search engines and AI models, and we discussed which AI bots to allow on the website and which to block in our article about AI Robots on Your Website.
How much will it cost you if you leave it to an agency?
Some good news to wrap things up: for a typical e-shop or corporate website, this isn’t a major audit, but rather a minor adjustment to the chatbot template and a single page containing the terms and conditions. If your website is managed on a monthly basis by a reputable agency, this modification will likely be included in your flat-rate fee. As a one-time modification outside of the management contract, it usually costs a few thousand crowns, depending on how many chatbots and templates you’re modifying.
A more comprehensive AI Act audit for the entire company—covering internal processes, documentation of AI use, and team training—is a separate and significantly more extensive service currently offered by specialized firms focused on AI governance. For a typical company website, however, the steps we described above are usually sufficient.
Frequently Asked Questions
Does the chatbot on my website have to say that it's AI?
Yes, as long as the chatbot engages in a casual conversation and it isn’t obvious from the context that it’s a machine. You must provide this information clearly, no later than during the first interaction; a sentence hidden in the terms and conditions isn’t enough.
What happens if I don't comply with the AI Act?
A violation of the transparency obligations under Article 50 may result in a fine of up to 15,000,000 euros or 3 % of the company’s global turnover, whichever is higher. In the Czech Republic, oversight is provided by the Czech Telecommunications Office (ČTÚ), the Czech National Bank (ČNB), and the Office for Personal Data Protection (ÚOOÚ); however, these authorities are currently focusing primarily on education.
Has the AI Act been postponed, or not?
Only partially. The AI Omnibus Act of July 2026 postponed the rules for high-risk systems until the end of 2027 and 2028. Transparency requirements for chatbots and AI-generated content will take effect as planned on August 2, 2026.
Does the AI Act apply to small businesses and self-employed individuals as well?
Yes. Unlike the Web Accessibility Act, there is no exemption for micro-enterprises here. The size of the company only affects the amount of any potential fine, not whether the requirement applies.
Do I have to label text on my blog that was written using AI?
Only if it is not subject to human editorial review and serves to inform the public about matters of public interest. An article that undergoes human editorial review is not subject to this obligation.
Who oversees compliance with the AI Act in the Czech Republic?
As of August 2, 2026, oversight has been divided among three agencies: the Czech Telecommunications Office as the lead agency, the Czech National Bank for the financial sector, and the Office for Personal Data Protection in cases where AI processes personal data.
What to Take Away From This
The AI Act isn’t a threat that will materialize sometime in the future—it’s been in effect for over two months and applies to every website with a chatbot or AI-generated content, regardless of the company’s size. In practical terms, this means three things: updating the chatbot’s welcome message and label, reviewing the terms of use, and knowing when you must disclose AI-generated content. None of these steps are expensive or complicated, as long as you don’t neglect them. Apadore handles these adjustments as part of its website and e-store management services.
Not sure if your chatbot is doing what it's supposed to? Contact us and we'll go through it with you; we'll tell you straight up what needs to be changed and what doesn't.
Would you like to have a chatbot? all right?
We'll review your website and suggest changes that will bring it into compliance with the new rules.
The AI Act isn’t the only European regulation that will affect your website in the near future. The Council of the EU has approved a position on the Digital Omnibus, a proposal to simplify cookie consent. We describe exactly what the proposal changes regarding cookies on websites in this article Digital Omnibus: What the Proposed Cookie Simplification Changes (and Doesn't Change) on Your Website.
The AI Act isn’t the only new requirement changing the way websites look this year. Starting January 1, 2027, e-commerce sites and online services will be required to include a “cancel order” button. To find out exactly what it must be able to do and how to test it, see Article on the "Cancel Order" button.




