A store adds a chat assistant that answers sizing questions, generates lifestyle photos for a new collection and drafts product descriptions with a language model. A year ago, none of that came with a legal question. In the EU, since August 2, 2026, each of them can.
Article 50 of the EU AI Act sets transparency rules for AI that talks to people and for content AI produces. In this article, we go through the parts that touch online stores, based on the European Commission's guidance, and end with a short checklist.
This article is for information only and is not legal advice. Whether a rule applies depends on your role, your system and your market; check your own case with a lawyer.
When it applies
According to the European Commission, Article 50 applies from August 2, 2026. There is one limited grace period: for AI systems placed on the market before that date, the obligation to mark AI-generated content applies from December 2, 2026. Content generated before August 2, 2026 doesn't have to be labeled retroactively, although the Commission encourages it.
Fines can reach €15 million or 3% of total worldwide turnover, with proportionality for small and medium-sized companies.
Provider or deployer?
The Act splits obligations by role.
- A provider develops an AI system, or has it developed, and puts it on the market or into service under its own name or trademark. The Commission says providers outside the EU are covered if their system's output is used in the EU.
- A deployer uses an AI system under its own authority in a professional activity.
A store that installs a vendor's chat widget is usually a deployer of that widget. A store that builds its own assistant and runs it under its own brand may be a provider. The obligations differ, so it's worth knowing which you are for each tool.
1. Assistants that talk to customers
Article 50(1) requires providers of AI systems that interact directly with people to design them so that people are informed they are interacting with AI, unless that is obvious. The Commission's guidance lists four conditions: it is an AI system, it is built for a genuine two-way exchange, the AI communicates directly rather than through a human, and the other side is a natural person.
The notice must come from the start of the first interaction, be clear and distinguishable and meet accessibility requirements. The "obvious" exception is to be read narrowly.
What to do: if you run a shopping assistant, make sure its first message or its window says plainly that it is an AI assistant. If a vendor supplies it, check that this is built in, and that your own changes to the greeting don't remove it. We looked at shopping assistants on store sites in Copilot Checkout and Brand Agents.
2. AI-generated images that look real
Two rules meet here.
- Marking (Article 50(2)). Providers of generative AI systems must mark synthetic images, video, audio and text in a machine-readable way, so the output can be detected as AI-generated. This falls on the tool's provider, not on the store, and it doesn't apply when the AI only performs standard editing.
- Deepfake disclosure (Article 50(4)). Deployers must disclose deepfakes: AI-generated or manipulated content that resembles existing persons, objects, places or events and would falsely appear authentic. The disclosure must be visible to the person at first exposure; the Commission says a deployer can't rely on the provider's machine-readable mark alone.
Whether a product image is a deepfake depends on context. The Commission's guidance looks at the level of resemblance, the message and what the audience expects. A clearly stylized illustration is different from a photorealistic image of a model wearing your product who doesn't exist, or of a product shown in a way it doesn't look in reality.
What to do: keep a record of which images are AI-generated. For photorealistic images of people or of your products, consider a visible note. Don't strip the marks your tools add. On this site, every insight illustration carries a visible note that it was generated with AI.
3. AI-written text
Article 50(4) also requires deployers to label AI-generated or manipulated text published to inform the public on matters of public interest. The Commission lists examples such as politics, public health, the environment and consumer safety. Text that has gone through human review or editorial control, with a person holding legal responsibility for it, doesn't need the label.
Ordinary product descriptions are generally not published to inform the public on matters of public interest. Safety information, health claims or guides on public topics may be closer to that line.
What to do: have a person review AI-drafted text before publishing, especially safety, health and regulatory content. A review step is good practice anyway: an AI-written description that invents a material or a certification is a problem for your customers before it is a legal one.
For stores outside the EU
The Commission's guidance says providers outside the EU are covered when their system's output is used in the EU. If you sell to EU customers from Turkey, the UK or elsewhere, assume these rules can matter to the tools you use with those customers, and check your position with a lawyer.
A short checklist
- List every AI tool that faces customers or produces published content: chat assistants, image generators, copy tools.
- For each, note whether you are the provider or the deployer.
- Make sure customer-facing assistants say they are AI from the first message.
- Keep a record of AI-generated images; add visible notes to photorealistic ones.
- Leave machine-readable marks from your tools in place.
- Put a human review step on AI-drafted text, and keep a record of who approved what.
The bottom line
For most online stores, Article 50 comes down to three habits: an assistant that says it is AI, honesty about images that look real but aren't, and a person who reviews AI-written text before it goes live. None of these conflicts with selling well. Shoppers and agents both trust a store more when what it says about itself is true.
Specoria doesn't assess legal compliance. We check whether the facts a store publishes about its products and policies are readable and consistent, which is the same discipline that keeps an AI assistant from repeating something untrue. Start with a free readiness test.
Sources
- European Commission, Transparency obligations under Article 50 of the AI Act, last updated July 24, 2026.
- Specoria Insights, Copilot Checkout and Brand Agents