Chapter 6 · Mandatory from 2 Aug 2026
AI in your shop: what applies since 2 August
Since 2 August 2026, the transparency rules of the EU AI Act apply. For online shops, that mainly means two things: a chatbot has to identify itself as AI, and deceptively real AI images and videos need a visible notice. Here is which role you have, what you need to do, and what, despite many posts saying otherwise, does not apply to you at all.
Last checked: · A reference, not legal advice.
The essentials
- Since
- 2 August 2026 (Article 50 AI Act). Content created before that does not need to be labelled retroactively.
- Chatbot
- People chatting with an AI must be able to tell from the first message. That is the provider’s duty. If you built the bot yourself or had it built, that is you.
- Images and video
- Deceptively real AI content (deepfakes) needs a visible notice. That applies to you when you create it for your business.
- Product copy
- According to the Commission, product descriptions and ad copy written with AI do not need an AI notice.
- 2 December 2026
- Deadline for invisible, machine-readable marking by AI generators already on the market before August. That is for the makers of those tools, not for your shop.
- Enforcement
- National market surveillance authorities; in Germany, the Federal Network Agency. Fines up to €15 million or 3 % of worldwide annual turnover; for SMEs, whichever is lower.
Provider or deployer: which role you have
The AI Act assigns duties by role. A provider develops an AI system or has it developed and places it on the market or puts it into service under its own name. A deployer uses an AI system under its own authority in a professional context. A shop is almost always a deployer, sometimes a provider as well.
- You use a ready-made chat tool or image generator as the maker supplies it: you are a deployer. The chatbot notice and the marking of images are the maker’s duty.
- You built your own chatbot or had an agency build it, and you run it in your shop under your name: according to the Commission, you are then a provider and must make sure the notice is there yourself.
- You take someone else’s AI system, modify it substantially, for example with your own training data, and run it under your name: you become a provider too.
- Your staff, freelancers or contractors work with AI on your behalf and under your control: you remain the deployer, not them.
- You commission an agency to make an ad and do not decide whether or how it uses AI: according to the Commission, you are not a deployer. The agency is.
- You create content as a creator regularly and for money: that is professional, not personal use. You are a deployer.
Open or disputed
- Where exactly a heavily customised chat tool with your own name, instructions and data turns you from deployer into provider is left open by the regulation and the guidelines. When in doubt, make sure the notice is there yourself.
The chatbot notice
The Commission explicitly lists chatbots in e-commerce and customer support as cases where the notice is required. It reads the exception “it is obvious anyway” narrowly: it only applies when an average person has almost no doubt left. According to the guidelines, a chat window in a shop does not qualify.
According to the Commission, recommendations (“You might also like”), automatic translation and search that does not generate content are among the things not covered. If AI only helps your support staff write their replies, the AI does not talk to customers directly, so the chatbot notice does not apply.
Before the first reply
The notice has to be there at the latest at the first interaction, clear and distinguishable. For example above the input field: “You are chatting with our AI assistant.”
Accessible
The regulation requires the notice to meet accessibility requirements. So real text that screen readers can read, not just an icon.
Humans and AI mixed
If sometimes a human answers and sometimes the AI, the AI answers have to be recognisable. Exception: a human properly reviews the answer and sends it themselves.
AI agents
If an AI acts on your behalf, for example booking or negotiating, the guidelines say it should disclose that it is an AI and on whose behalf it acts.
AI images and video: when a notice is required
Not every AI image needs a label, only deepfakes. The AI Act means AI content that resembles real persons, objects, places or events and would falsely appear authentic. Objects are explicitly included, so your product counts too.
The Commission’s guidelines give advertising examples. A deepfake is an AI product image that shows the product differently, more appealing or of higher quality than it really is. So is a realistic AI influencer testing a real product, or a teleshopping-style video with synthetic customers. Not a deepfake: your real product in front of an AI background, as long as the image does not mislead about the product. Colour correction, a background swapped for purely aesthetic reasons, arranging real products or upscaling an image have, according to the Commission, little effect on how authentic the content appears.
Where a label is required, it has to be visible or audible, at the latest when someone first sees the content. The invisible marking in the file is not enough, the Commission says explicitly. And importantly: the notice does not make a misleading product image legal. If an image shows your product as better than it is, that is still misleading advertising.
AI text in your shop
For text, there is only one labelling duty: published text meant to inform the public on matters of public interest, such as health, the environment or consumer safety. The Commission explicitly names product descriptions and ad copy as an example that is not covered, except for claims related to health, consumer safety or sustainability, for example.
The duty does not apply when a knowledgeable person reviews the content and someone holds editorial responsibility. According to the Commission, spell-checking does not count as review.
Open or disputed
- How far the exception for health, consumer safety and sustainability claims reaches in product copy is not explained further in the guidelines. If your shop publishes AI-written guides on health or the environment, have them reviewed properly or label them.
What does not apply to you right now
- Machine-readable marking (Article 50(2)) is a duty of the AI tool providers, not of the shops using them. For tools already on the market before 2 August 2026, the Digital Omnibus moved it to 2 December 2026.
- Content created before 2 August 2026 does not need to be labelled retroactively. The Commission does encourage it.
- The strict rules for high-risk AI, which include AI for credit scoring of individuals, only apply from 2 December 2027 under the Digital Omnibus.
- The Digital Omnibus softened the AI literacy duty (Article 4): providers and deployers must take measures to support the AI literacy of their staff. Nobody has to guarantee a specific level.
- Purely personal use of AI is outside the AI Act. That ends once you earn money with it on a regular basis.
Open or disputed
- Whether competitors in Germany can also send cease-and-desist letters under the UWG for a missing AI notice has not been decided by any court I could find.
- The Commission’s code of practice on labelling AI content is voluntary. Those who do not sign it have to show compliance in other ways.
Sources
- Regulation (EU) 2024/1689 (AI Act), EUR-Lex
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex
- European Commission: guidelines on transparency obligations under Article 50
- European Commission: questions and answers on Article 50
- Germany: AI Market Surveillance and Innovation Act (KI-MIG, German)
Questions about this chapter
Do I have to label every AI image in my shop?
No. The notice is required for deepfakes, meaning AI content that appears deceptively real. According to the Commission, an AI background behind your real product or colour correction usually does not count. An AI product image that makes the item look better than it is does.
I use a ready-made chat plugin. Do I need to do anything?
The notice is the plugin maker’s duty. Still check that it actually appears in your shop before anyone sends the first message. If it is missing, switch it on or ask the maker.
Is the invisible watermark my image generator adds enough?
Not for deepfakes. The Commission says explicitly that as a deployer you cannot rely on machine-readable marking. The notice has to be visible or audible to people without special tools.
Do I have to label old images retroactively?
No. According to the Commission, content from before 2 August 2026 does not need to be labelled retroactively. It does recommend it where possible.
What about AI avatars and AI influencers in my videos?
If the person looks real, it is a deepfake and needs a visible notice. The guidelines name a realistic AI influencer testing a real product as an explicit example. The lighter rule for art or satire does not apply to that kind of advertising.
Who enforces this and what are the risks?
National market surveillance authorities; in Germany, the Federal Network Agency since 29 July 2026. The AI Act allows fines up to €15 million or 3 % of worldwide annual turnover; for SMEs, whichever amount is lower.
Is this legal advice?
No. It is the current state with sources and dates. For your specific case, ask a lawyer or your trade association.
Related rules
- Chapter 7Shop accessibility Online shops selling to consumers must work for everyone, including keyboard and screen reader users. Only microenterprises are exempt.
- Chapter 3Green claims “Climate neutral” based on offsetting is banned, and so are unproven “eco-friendly” claims and self-made labels.
- Chapter 8Product safety (GPSR) Every listing needs the manufacturer’s address, a product picture and identifier, and the safety warnings.