On August 2, 2026, the transparency obligations in Article 50 of the EU AI Act became enforceable, and most ecommerce and marketing teams have not looked at what it actually requires on a website. It is not a broad AI ban and it is not about your ad targeting. It is a narrow set of disclosure rules for chatbots, AI shopping assistants, product recommendation engines and AI-generated content, and it applies to any business whose AI output reaches an EU visitor, wherever that business is based. Here is what it actually says, who it hits, and what it does to your conversion rate if you handle it badly.
Since August 2, 2026, any AI system that talks to a person or generates content for them, chatbots, AI shopping assistants and AI-powered product recommendations included, must clearly disclose that it is AI before or at first interaction if that is not already obvious. AI-generated images, video, audio and text need machine-readable marking, with existing systems given until December 2, 2026 for that specific piece. It applies to any business whose AI output reaches an EU visitor regardless of where the business is based, and fines run up to 15 million euros or 3% of global turnover.
What does Article 50 actually require?
The AI Act's transparency obligations cover four areas: disclosure when a person is interacting directly with an AI system such as a chatbot, voice assistant or AI agent; marking of AI-generated or AI-manipulated audio, image, video and text so it can be machine-detected; disclosure when emotion recognition or biometric categorisation systems are used on someone; and labelling of deepfakes and AI-generated text published on matters of public interest. For a typical ecommerce or B2B site, the first two are the ones that matter. The obligations became applicable on August 2, 2026 and apply immediately to every in-scope system regardless of when it was built or deployed, with enforcement now sitting with the EU's AI Office and the relevant authority in each member state.
| Obligation | What it covers | Deadline |
|---|---|---|
| Interaction disclosure | Chatbots, voice assistants, AI shopping agents, AI product recommendations | Live since August 2, 2026, no grace period |
| Content marking | AI-generated or manipulated images, audio, video, text | New systems: live now. Existing systems: until December 2, 2026 |
| Emotion/biometric disclosure | Systems inferring emotion or categorising people by biometric data | Live since August 2, 2026 |
| Deepfake and public-interest labelling | Synthetic media and AI text on public interest topics | Live since August 2, 2026 |
Does your AI shopping assistant or recommendation widget need a label?
In practice, yes, for most implementations. AI-driven recommendation apps, the kind built into Shopify's Search & Discovery, Nosto, or a similar engine, count as an AI system interacting with the customer under Article 50(1), and legal guidance circulating since the rollout treats a simple visible line such as "Recommendations powered by AI" near the module as sufficient. The requirement is that the disclosure be clear, distinguishable and delivered no later than the customer's first exposure to it, not that it follow a specific format or wording. The same logic covers an on-site AI chat widget or an AI shopping assistant: if it is not obvious to a reasonable visitor that they are talking to AI, you disclose it.
Do not bury the disclosure in a privacy policy nobody reads. Put a one-line label directly on the recommendation module or at the top of the chat window, test two or three short phrasings for tone, and measure whether disclosure actually hurts trust or, more often, makes no measurable difference once the wording is calm and not alarmist. Treat it the same way you would test any other piece of on-page copy.
Does this apply if my business isn't based in the EU?
Yes, and this is the part most US small businesses miss. Article 3(4) applies the AI Act extraterritorially: it covers any AI system whose output is used in the EU, regardless of where the provider or deployer is located. A US-based Shopify store selling to customers in Germany or France is in scope for the AI systems it deploys to those visitors, even if the business has no EU office, staff or legal entity. If your conversion optimization stack includes any EU traffic and any AI-driven personalization, chat or content generation, you are covered.
What is the actual penalty exposure?
Fines for non-compliance with the transparency obligations can reach up to 15 million euros or 3% of worldwide annual turnover, whichever is higher, with a lower cap of up to 750,000 euros for EU institutions, bodies and agencies. The Act does build in proportionality for small and medium-sized businesses when penalties are assessed, but proportionality is not the same as exemption, and there is no small business carve-out from the disclosure requirement itself. For most sites reading this, the realistic risk in year one is less about a fine and more about a complaint or a competitor report triggering a look at your site, which is a good reason to fix the obvious gaps now rather than wait for enforcement patterns to become clear.
What about AI-generated product photography and ad creative?
This is where the marking obligation bites. Machine-readable marking is required for AI-generated or AI-manipulated audio, image, video and text, so that the content can be identified as synthetic. If you generate product lifestyle shots, model images or video ad creative with AI tools, new generation from here forward needs this marking built in wherever the tool supports it. Providers whose generative systems were already on the market before August 2, 2026 have a transition window running to December 2, 2026 specifically for the marking and detection piece, but the interaction-disclosure requirements around chat and recommendations carry no such grace period.
A practical audit for the next two weeks
- List every AI-driven touchpoint on your site: chat widgets, shopping assistants, product recommendation modules, quiz-style personalization, and any AI-generated imagery or video in active use.
- For each one, check whether a reasonable visitor would already know it is AI. If not, add a short, visible disclosure at first exposure.
- Check whether any of that AI output reaches EU visitors, by traffic, currency options, or shipping destinations. If it does, treat the whole flow as in scope.
- Ask your ecommerce platform or app vendor directly whether their AI features already ship with a compliant disclosure, since several major recommendation apps have added one since August.
- Note which AI-generated content assets predate August 2, 2026, since those get the marking grace period through December 2; anything generated after that date should already carry it if your tool supports it.
None of this is a reason to rip out AI personalization or an assistant that is genuinely lifting conversion. It is a reason to add a line of copy most teams have been putting off, and to do it before a regulator or a competitor does it for you.
Frequently asked questions
What is the EU AI Act's Article 50 and when did it take effect?
Article 50 sets transparency obligations for AI systems and became applicable on August 2, 2026. It requires disclosure when a person interacts with AI, when content is AI-generated or manipulated, and when emotion recognition or biometric categorisation is used. It applies immediately to all in-scope systems, regardless of when they were placed on the market.
Does an on-site AI chatbot or shopping assistant need a disclosure now?
Yes, if it is not already obvious to users that they are talking to AI. Article 50(1) requires providers of AI systems that interact directly with individuals, including chatbots, voice assistants and AI shopping agents, to disclose that clearly at or before the first interaction.
Do AI-powered product recommendations need a disclosure too?
In most reasonable interpretations of Article 50(1), yes. AI-driven recommendation apps count as an AI system interacting with the customer. A simple, visible line such as "Recommendations powered by AI" near the recommendation module is generally treated as sufficient, delivered no later than the customer's first exposure to it.
Does this apply to a US business that just sells to EU customers?
Yes. Article 3(4) of the AI Act applies extraterritorially: it covers any AI system whose output is used in the EU, regardless of where the provider or deployer is based. A US-based Shopify store selling to German or French customers is in scope for the AI systems it deploys to them.
What happens if a business does not comply?
Fines for non-compliance with the AI Act's transparency obligations can reach up to 15 million euros or 3% of worldwide annual turnover, whichever is higher, with proportionality provisions for small and medium-sized businesses. Enforcement sits with the EU AI Office and national authorities from August 2, 2026 onward.
Is there a grace period for AI-generated content already on my site?
A limited one, and only for the marking and detection obligation on generative AI systems already on the market before August 2, 2026. Those providers have until December 2, 2026 to add machine-readable marking. The interaction-disclosure requirement for chatbots and recommendation engines has no such grace period.
The takeaway
Article 50 is narrower and more manageable than the phrase "EU AI Act" makes it sound. If you run an AI chatbot, an AI shopping assistant or AI-driven product recommendations and any of your traffic is European, add a short, honest disclosure this month. Test the wording like you would any other line of copy, do not treat it as a reason to switch the feature off, and check with your platform or app vendor whether they have already shipped a compliant default. The businesses that get caught out here are not the ones testing disclosure copy. They are the ones who never looked.