EU AI Act: Labeling Obligations for Chatbots in Online Shops

1. What is the EU AI Act labeling obligation for chatbots?

The EU AI Act, formally Regulation (EU) 2024/1689, is the first comprehensive European law regulating artificial intelligence. It entered into force on 1 August 2024 and governs how AI systems may be developed, offered and used within the European Union, phased in over several years. For online shops, one rule is especially relevant: a duty the text itself calls a transparency obligation, and which is often referred to in plain language as the labeling obligation for chatbots.

This rule, known among specialists as Article 50 of the AI Act, applies to AI systems intended to interact directly with natural persons, for example a chat window on a shop website. It requires that people be informed that they are communicating with an AI system, not a human staff member. An exception applies only where this is already obvious from the circumstances and context of use, which is rarely the case in the everyday setting of an online shop.

What this covers is above all the classic scenario: a customer opens a chat window in your shop, types a question about shipping costs or returns, and gets an immediate answer. Whether a simple rule set or a modern language model sits behind it partly decides whether and how the obligation applies. The underlying idea of the legislator matters here: customers should be able to decide for themselves whether they want to talk to a machine or a human, and for that they need a clear, visible piece of information, not a hidden footnote.

Clearly distinct from this chatbot rule are pure product recommendations, that is, features like "customers also bought" or "you might also like". These systems generally do not trigger the same kind of obligation, more on that later in this article, so that by the end you know exactly where action is needed and where it is not.

2. Since when does it apply, when does it become relevant?

The EU AI Act has been in force since 1 August 2024, but it becomes applicable in several stages so that businesses and authorities have time to prepare. Prohibited AI practices, such as certain forms of manipulative behavioral techniques, already applied from February 2025. Obligations for providers of particularly capable, general-purpose AI models followed from August 2025.

What matters most for online shops: most of the remaining obligations of the AI Act, including precisely the transparency obligation for chatbots under Article 50, become applicable from 2 August 2026. In concrete terms, this means that by the time you are reading this article, this deadline has either just passed or is imminent, depending on exactly when you read it. For practical planning, you should therefore treat the obligation as already fully in effect, since waiting for a generous grace period is not the safer option here.

One point worth stating honestly: responsibility for monitoring the AI Act lies with national market surveillance authorities, and in several EU member states these authorities and their exact responsibilities were, at the time this article was published, still being organized or formally designated. Details of actual enforcement practice are therefore likely to keep settling in over the coming months. This does not change the fact that the substantive labeling obligation itself applies from the date mentioned, regardless of how quickly individual authorities fully ramp up their oversight activity.

For you as a shop owner, this means in very practical terms: do not wait for someone to officially knock on your door before introducing labeling. Anyone who knows their chat widget provider and what runs behind the scenes can usually implement labeling within a few days, often with little effort.

3. Who is affected, does this apply to small shops too?

Yes, explicitly also to small and side-business shops. The transparency obligation itself carries no exemption for small businesses, unlike some other rules where turnover thresholds or headcount play a role. What matters is solely whether you use an AI system intended to interact directly with people, not the size of your business or your annual turnover.

This applies even if you did not build the chatbot yourself but merely integrated a ready-made plugin or a third-party service into your shop. From the customer's perspective, it does not matter who built the technology, only that it runs in your shop and talks to your customers. As the operator of the shop, you therefore carry the responsibility for ensuring clear information is in place, even if the underlying AI technology comes from an external provider.

Importantly, though, the principle of proportionality applies: the effort expected by the legislator scales with the complexity and risk of the particular AI system. A small shop with a simple, clearly recognizable chat assistant does not need to build an elaborate compliance department. A visible, understandable notice in the right place is usually already enough to meet the obligation, with no bureaucratic overhead.

4. What you specifically need to do

The first and most important step is a short inventory: which chat, assistant or recommendation tools currently run in your shop, and does artificial intelligence, such as a language model, actually operate behind any of them? Many plugin providers state this in their own product description, though sometimes only in the fine print of the technical documentation, so a second look is worthwhile.

Once it is established that an AI system is involved that interacts directly with your customers, clear, well-visible labeling is needed. In practice, a short notice directly on the chat window works well, for example a label reading "AI Assistant", combined with a brief introductory message at the start of every conversation that explicitly states an artificial intelligence, not a human, is answering.

Also avoid a design that creates the impression of a real staff member, for example through a made-up first and last name together with a portrait photo, without it also being clear that this is an AI. A name alone is usually not problematic as long as the AI labeling remains clearly recognizable and is not obscured by the rest of the design.

Also review every third-party plugin in your shop, not just the obvious chatbot, since recommendation widgets or support tools can also use AI functionality behind the scenes without it being obvious at first glance. Finally, keep a short, informal internal note of which AI tools you use and where, which makes later reviews and updates considerably easier should tools or providers ever change.

6. Fines and risks of non-compliance

The EU AI Act provides for a tiered system of fines whose exact amount depends on the type of violation. Without stating a specific figure here, since it can vary strongly by violation category, it is fair to say: sensitive fines are at stake, oriented, depending on the severity of the violation, around a percentage of a company's worldwide annual turnover or a high fixed amount in the millions, whichever is higher. For the transparency obligation around chatbots, the lower tiers of the fine framework are more likely to be relevant in practice than the maximum rates reserved for the most serious violations.

For most small and medium-sized shops, another risk matters more than the abstract figure anyway: the loss of customer trust when it becomes public that an unlabeled AI chat posed as a supposed human being. Such cases often spread quickly on social media and can damage a shop's reputation more lastingly than a one-off fine, precisely because trust plays a central role in purchase decisions in online retail.

Since the responsible market surveillance authorities in the member states are only gradually building up their review and enforcement practice, it is not yet fully foreseeable how strictly and how quickly individual violations will actually be pursued in practice. This is, however, no reason to postpone labeling, since the effort for a clean implementation is manageable, while the uncertainty around future enforcement practice actually argues for an early, cautious approach.

7. Common misconceptions

A widespread misconception goes: "I don't have 'real' AI, just a simple chatbot plugin, this doesn't affect me." Whether the obligation applies, however, does not depend on how sophisticated or impressive a system feels, but on the fairly broad legal definition of an AI system used in the AI Act. Even a seemingly simple plugin can fall under it if it actually uses a learning or language-model-based system behind the scenes, rather than a purely rigid rule set.

A second, equally widespread misconception goes: "product recommendations always need an AI label." In this blanket form, that is usually overstated: classic recommendation features such as "customers also bought" generally do not count as high-risk AI and normally do not trigger an explicit labeling obligation, as long as they do not generate independent, synthetic content or interact conversationally with the person. Still, a short, voluntary transparency note can benefit customer trust here too, even where it is not legally required.

A third misconception concerns the permanence of labeling: "labeling once is enough forever." If a chat widget is later extended with new features, for example a more capable language model or additional conversational options, the labeling should be reviewed alongside it to make sure it stays accurate and complete. A brief review, for instance once a year or with every major change to the chat tool, is usually enough for this.

8. What Mironsoft can take care of for you

We know that terms like EU AI Act, transparency obligation or Article 50 tend to create more confusion than clarity for shop owners without a technical background, especially since new European rules are often phrased in abstract terms. Mironsoft takes care of the practical side for you: together with you, we look at which chat, assistant and recommendation tools are actually in use in your shop, and clarify where genuine AI is running behind any of them.

We then make sure clear, well-visible labeling appears in the right places, matched to your shop's design, without it looking like a bolted-on warning notice. We also keep a short record of which AI tools you use, so that if in doubt you can quickly retrace what was set up and when, without any searching.

Since the interpretation of the AI Act and the practice of the responsible authorities are likely to keep evolving over the coming months, we keep an eye on relevant changes for you and get in touch proactively if anything about your labeling should be adjusted. This way you do not have to work through European regulatory texts yourself, but have a single point of contact who tells you when and where action is genuinely needed, without unnecessary rush or alarm.

9. Summary

Article 50 of the EU AI Act requires that customers can clearly recognize when they are talking to an AI chatbot or AI shopping assistant rather than a human. This transparency obligation becomes applicable from 2 August 2026 and applies regardless of the size of your shop, as long as you use an AI system that interacts directly with people. Classic product recommendations are generally not subject to the same kind of obligation, unless they generate independent, synthetic content or interact with the person conversationally.

Implementation is usually straightforward: a visible label, a short introductory chat message and an honest look at the tools you use are enough in the vast majority of cases. This article offers a general overview and does not replace individual legal advice for your specific case.

10. FAQ: EU AI Act and Labeling Obligations for Chatbots

This article offers a general overview of the EU AI Act labeling obligation and does not replace individual legal advice for your specific case.