Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026. For a normal online shop the answer is narrower than the headlines suggest. Product copy written with an AI writing tool usually needs no label. An AI generated product image that could mislead a shopper about the real product does need one. And a chatbot has to make clear that it is a chatbot.
The duties in Article 50 are split between the company that builds an AI system and the company that uses it. As a shop owner you are almost always in the second group, so only part of the article lands on your side. Below is which part that is, and what a compliant disclosure looks like in a shop.
What Article 50 actually says
Article 50 sets out four separate duties, and it matters which one is yours.
- Article 50(1), providers: an AI system meant to interact directly with people has to be designed so that those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well informed, observant and circumspect person.
- Article 50(2), providers: systems generating synthetic audio, image, video or text have to mark their output in a machine readable format so it is detectable as artificially generated or manipulated.
- Article 50(3), deployers: emotion recognition and biometric categorisation systems require you to inform the people exposed to them.
- Article 50(4), deployers: deep fakes have to be disclosed, and AI generated text published to inform the public on matters of public interest has to be disclosed unless a person reviewed it and carries editorial responsibility for it.
A provider develops an AI system and places it on the market under its own name or trademark. A deployer uses a system under its own authority. If you licence a chatbot or an image generator and run it on your shop, you are the deployer, and the duty that concerns you is Article 50(4). The Commission guidelines confirm that employees and freelancers acting on your instructions do not become separate deployers, the company stays the deployer. Whichever duty applies, Article 50(5) requires the information to be clear and distinguishable at the latest at the first interaction or exposure.
Which shop content needs a label and which does not
Four content types cover almost everything a shop publishes, and only two of them carry a duty.
Product descriptions and marketing copy
The Commission guidelines published on 20 July 2026 are unusually concrete here. They are not binding, and only the Court of Justice can settle what Article 50 means, but they are the clearest signal available on how the Commission expects the article to be read. They list “AI-manipulated text that is part of a company’s advertisement or product descriptions” as an example of text that falls outside Article 50(4), because it is not published to inform the public on matters of public interest. A generated bullet list for a running shoe therefore needs no AI label.
One bracket in that example is worth reading twice. It excludes copy carrying claims on health, consumer safety or sustainability. A supplement description with health claims sits much closer to the public interest category. If your catalogue is regulated in that way, route generated copy through a named person who reviews and signs it off, which is also the exemption the AI Act itself offers.
Product and campaign images
This is where most shops are actually exposed. Article 50(4) applies to deep fakes, and Article 3(60) defines a deep fake as AI generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic or truthful.
The guidelines spell out what that means in a shop. An AI generated image of a product in an advertisement or on packaging counts as a deep fake if it can mislead the audience about the actual product appearance, characteristics or use, for example by making the product look different from the real item, more appealing or of better quality. A real product photographed against an AI generated background is listed as not a deep fake, as long as the ad does not mislead about the product itself. Colour correction, noise reduction, background replacement for clearly aesthetic purposes and rescaling are listed as changes that normally have only a minor impact, and the guidelines add that this depends on the context of the individual case.
The practical line: a lifestyle scene built around a real product shot is fine. A product visual invented by a model, where a customer would assume they are looking at a photograph of the item they are buying, needs a visible disclosure.
Chatbots and AI assistants
The chatbot duty in Article 50(1) sits with the provider of the system, which is normally your vendor. Your practical job is not to undo it. The guidelines warn that a human sounding voice, a human profile picture and advanced personalisation all reduce how obvious the artificial nature of an interaction is, and that general awareness that chatbots exist does not mean people recognise one in a given conversation. The obviousness exception has to be read restrictively.
So if you rename the assistant, give it a portrait and a first name and switch off the vendor’s opening disclosure, you have made the interaction less obvious, not more. And where a system is put into service under your own name or trademark, you can end up in the provider role yourself, with the design duty attached. Settle that with your vendor before launch.
Customer reviews
Reviews are not an AI Act problem in the first place, they are an unfair commercial practices problem, and those rules are older and harder. Annex I of Directive 2005/29/EC lists submitting or commissioning false consumer reviews, and misrepresenting consumer reviews, as practices prohibited in all circumstances (point 23c). Point 23b catches the neighbouring case: stating that reviews come from consumers who have actually used or bought the product, without taking reasonable and proportionate steps to check that they do. No balancing test, and no label that repairs either of them.
How to put the disclosure in place
Where a disclosure is required, the guidelines are clear that it has to be perceivable by a person without special tools. Deployers cannot rely on the machine readable marking that the provider embeds under Article 50(2), because that marking is not visible to the shopper.
What works in a shop:
- A visible caption or badge on the image or directly next to it, in the language of the page.
- A first message in the chat stating that the assistant runs on AI, combined with a small persistent label near the input field.
- Plain wording. The guidelines suggest phrasing at the level of “You are interacting with an AI system”.
What does not work: burying the statement in your terms and conditions, in a linked document or in a URL. The guidelines name those as insufficient on their own. Article 50(5) also requires the disclosure to meet the applicable accessibility requirements, so a badge that exists only as a decorative image without a text alternative will not carry it.
Dates, penalties and what stays open
Article 50 has applied since 2 August 2026 under Article 113 of the AI Act. There is one narrow transition. Regulation (EU) 2026/1744, the Digital Omnibus on AI, gives providers of generative systems that were already on the market before 2 August 2026 until 2 December 2026 to bring the machine readable marking under Article 50(2) into line. That grandfathering does not touch the chatbot disclosure and it does not touch deployer duties, both of which applied from 2 August 2026. The same amendment postponed several high risk timelines into 2027 and 2028, and left Article 50 where it was.
Content generated before 2 August 2026 does not have to be labelled retroactively, but text generated before that date and published on or after it does.
Article 99(4)(g) puts infringements of Article 50 in the tier of up to 15 million euro or 3 percent of total worldwide annual turnover, whichever is higher. For small and medium sized enterprises Article 99(6) turns that into whichever is lower. Enforcement runs mainly through the national market surveillance authorities, and the open question is how strictly they will read the deep fake definition for ordinary product imagery. The Commission calls its guidelines a first interpretation and will revise them as practice develops, so this is a topic to revisit rather than to file away.
The AI Act does not replace the rules you already have
Article 50(6) states that the transparency duties are without prejudice to other transparency obligations in Union or national law. A product visual that flatters the item beyond recognition was already a misleading commercial practice before August 2026. Article 50 adds a labelling duty on top of that, it does not turn a misleading image into a lawful one.
That is the honest summary for most shops. The AI Act is a small and targeted addition, while the larger compliance risk still sits in consumer law that has been in force for years, from the Omnibus Directive to the pricing indication rules.
Key takeaways
- Article 50 of the EU AI Act has applied since 2 August 2026.
- Product descriptions written with AI generally need no label, unless the copy carries health, consumer safety or sustainability claims.
- An AI generated product image that could mislead about the real product counts as a deep fake and needs a visible disclosure from you as the deployer.
- Machine readable watermarks do not satisfy the deployer disclosure, and a note in the terms and conditions does not either.
- Fake or AI written customer reviews are prohibited outright under the unfair commercial practices blacklist, no label helps.
- Breaches of Article 50 carry fines of up to 15 million euro or 3 percent of worldwide annual turnover.
Read next: how AI is changing online shopping, staying compliant with the Omnibus Directive, or our reference projects.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 3(60), 50, 99 and 113, EUR-Lex (checked 22 August 2026).
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), new Article 111(4) of the AI Act and the amended Article 113, EUR-Lex (checked 22 August 2026).
- European Commission, Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, published 20 July 2026, European Commission (checked 22 August 2026).
- European Commission, Questions and answers on the transparency obligations under Article 50 of the AI Act, European Commission (checked 22 August 2026).
- Directive 2005/29/EC, Annex I points 23b and 23c, as inserted by Directive (EU) 2019/2161, EUR-Lex (checked 22 August 2026).
This article reflects the legal position on 22 August 2026 and is general information about how the rules affect an online shop. It is not legal advice on your individual case.
Get your shop reviewed against the EU AI Act transparency rules
Lars Henning
- 42 ARTICLES
- 42 ARTICLES
