Last updated: 22 September 2026
The EU AI Act reaches a Shopify store through Article 50, the transparency rules for chatbots and AI-generated content. They apply from 2 August 2026. On 20 July 2026 the European Commission published guidelines that answer the main question for stores: when does an AI product photo need a label?
Breaking Article 50 can lead to fines of up to 15 million euros or 3% of worldwide annual turnover, and for SMEs the lower of the two amounts applies (Article 99 of the AI Act). In the Netherlands, no regulator can issue those fines yet, because the Dutch implementing law is still a draft. The duties themselves already apply.
This is practical guidance, not legal advice.
What does the EU AI Act require from a Shopify store?
Article 50 of the AI Act splits the work between providers, who build an AI system or put it into service under their own name, and deployers, who use one professionally. A store that uses AI tools and apps is a deployer.
- Article 50(1), providers: chat systems must be designed so people know they are dealing with an AI, unless that is obvious.
- Article 50(2), providers: AI images, video, audio and text must carry a machine-readable marking.
- Article 50(4), deployers: deepfakes must be disclosed, and so must AI text published to inform the public on matters of public interest, unless it went through human review and someone holds editorial responsibility.
If you build your own chatbot on an AI model and run it on your site under your store name, you can become the provider of that chatbot under Article 3(3). Then the Article 50(1) design duty is yours.
The Act defines a deepfake as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (Article 3(60)).
Does the EU AI Act apply to stores outside the EU?
Yes, when shoppers in the EU see the output. Article 2(1) covers deployers established in the EU, and also providers and deployers in a third country where the output of the AI system is used in the EU. A UK or US store selling to EU shoppers follows the same Article 50 rules.
Do AI product photos need a label?
Yes, when the image looks real and could mislead a shopper about the product or the people in it. The Commission's Article 50 FAQ says simulated persons and objects count when they resemble something that exists or could plausibly exist. Stibbe notes that the guidelines include products a company may sell. According to Bird & Bird, their examples include an AI product image in an advert or on packaging that can mislead about how the product looks.
- Label it: a fully or largely AI-generated product image that looks like a real photo and shows the product differently, or better, than it is. Examples are a smoother fabric, a richer colour or a shape the product does not have.
- Label it: a photorealistic AI model wearing your product. The person does not have to exist. A model who could plausibly exist meets the Commission's test.
- No label: your real, unaltered product on an AI-generated background, as long as the scene does not mislead about the product, its properties or its use (Bird & Bird).
- No label: standard edits such as cropping, brightness and contrast, dust removal or cutting the product out onto a plain background, when the product itself stays unchanged (Löffel Abrar).
- No label: scenes no shopper would take as real. The definition only covers content that would falsely appear authentic, and the Commission FAQ says what the audience expects in that context counts.
The exception for evidently artistic or fictional work still requires a short disclosure, and Bird & Bird reports that the guidelines exclude a realistic synthetic influencer testing a real sponsored product.
How do I add an AI label in Shopify?
The label must be clear and distinguishable at the latest when the shopper first sees the image (Article 50(5)). The guidelines reject a notice hidden under menu layers (Löffel Abrar). One way to do it without an app:
- Go to Settings > Metafields and metaobjects > Products and add a single line text definition, for example "AI image notice".
- On each product with a realistic AI image, fill in the field with a line such as "Model image created with AI".
- Open Online Store > Themes > Customize, go to your product template and add a text block in the product information section, close to the images.
- Use the connect dynamic source icon on that block to connect it to your new metafield (Shopify Help Center).
If you sell through Google Shopping, keep the label off the image file. Google Merchant Center does not accept overlays or watermarks on product images, and all images created with generative AI must contain metadata that says so, such as the IPTC DigitalSourceType tag (Google Merchant Center image requirements). Do not strip it when you compress images. Shopify's image CDN removed this metadata from uploads until Shopify staff reported a fix on 27 July 2026 (Shopify developer forum). Re-upload the original files of AI images you added before that date.
That metadata does not replace the visible label. The Commission FAQ says deployers cannot rely on the provider's machine-readable marking to meet their own disclosure duty. The Commission's EU icons for AI-generated content are free and optional to use. The labelling duty applies with or without them.
Do AI images and videos in Meta and TikTok ads need a label?
The same Article 50 test applies to ads. TikTok's ad policy requires the AIGC label or your own clear disclaimer on content significantly modified by AI. Meta adds an "AI info" label to ads made or significantly edited with its own generative AI tools, in the three-dot menu or next to "Sponsored" when it shows a photorealistic AI person. A label in a menu may not count as clear under the guidelines, so put your own line in the ad text for a realistic AI model or product scene.
Does an AI chatbot in a Shopify store need a disclosure?
Yes. Shoppers must know they are talking to an AI from the start of the first interaction. Under Article 50(1) the design duty sits with the provider of the chat app, and the chat still runs on your store.
In a joint position of 2 October 2025, the ACM and the Dutch Data Protection Authority (AP) said businesses should tell people they are dealing with an AI, offer a way to reach a person, and keep control over the information the chatbot gives.
- Send a test message in your store chat. If the first reply or the chat header does not say it is an AI assistant, add that line to the greeting in the app settings.
- Add a visible way to reach a person, such as an email address or a "talk to our team" option.
- Read a sample of past answers each month. A wrong return term from a bot is still your statement.
Do AI-written product descriptions need a label?
Generally no. The Article 50(4) text duty covers text published to inform the public on matters of public interest. According to Bird & Bird, the guidelines place company advertising and product descriptions outside that duty, with an exception for claims about health, consumer safety or sustainability. For those claims, human review lifts the disclosure duty. The Commission FAQ defines that as a person with relevant knowledge of the subject who checks the substance of the text, and an editor who can approve, change or reject it. Sustainability wording also has its own new rules, covered in our guide to EU green claims rules for Shopify stores.
A wrong size or material remains misleading under consumer law either way, which our guide to product descriptions that sell covers. Blog posts on public-interest topics such as health need a disclosure unless they had that human review. Bird & Bird reports the label belongs at the top of the text or near the headline.
What about AI recommendations, search and pricing?
Product recommendations, on-site search and pricing tools are not among the Article 50 uses and are not on the high-risk list, so the AI Act adds no label for them. Consumer law still applies. The ACM requires stores to tell shoppers before the purchase when an offer is personalised based on their data.
Article 4, as rewritten by the AI Omnibus, requires deployers to take measures to support the AI literacy of staff who work with AI systems. It does not require any specific level.
What to label, per use
| What you use AI for | Label or disclose? | Applies since |
|---|---|---|
| AI chat or sales assistant on your site | Yes, the chat must say it is an AI at the first message | 2 August 2026 as a provider duty. ACM and AP expectation since 2 October 2025. |
| Realistic AI product image that shows the product differently or better than it is | Yes | 2 August 2026 |
| Photorealistic AI model or avatar, including one who does not exist | Yes | 2 August 2026 |
| Real, unaltered product on an AI background | No, unless the scene misleads about the product | No new duty |
| Cropping, light correction, dust removal, plain background | No | No new duty |
| Product descriptions and ad text | Generally no. Health, safety or sustainability claims need human review. | 2 August 2026 |
| Blog post on a public-interest topic without human editorial review | Yes. With review and editorial responsibility, no. | 2 August 2026 |
| AI-generated reviews posted as if written by customers | Not allowed. These count as fake reviews. | 28 May 2022 |
Did the AI Omnibus delay the labelling rules?
No, not for businesses that use AI. The AI Omnibus, Regulation (EU) 2026/1744, entered into force on 27 July 2026. It moved the high-risk rules to 2 December 2027 and 2 August 2028. It gave providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to add machine-readable marking (White & Case).
That grace period is for the makers of AI tools. Your duties as a deployer apply from 2 August 2026. The voluntary Code of Practice on marking and labelling of 10 June 2026 is written for providers and deployers alike.
Who enforces the AI Act in the Netherlands?
Under the draft Dutch law, the AP and the AFM. The draft Uitvoeringswet AI-verordening went to public consultation on 20 April 2026, which closed on 1 June 2026. Article 2.2 of the draft names the AP and the AFM as market surveillance authorities for Chapter IV of the AI Act, the chapter that holds Article 50 (Raad van State, 8 July 2026). The AP and the RDI (the Dutch digital infrastructure inspectorate) share the coordinating role.
The draft still needs formal advice from the Raad van State and a vote in both chambers of parliament. Until it is adopted, no Dutch regulator can fine anyone under the AI Act. Consumer law already applies. The ACM has enforced the ban on fake reviews since 28 May 2022, and reviews also shape which brands AI assistants recommend.
Frequently asked questions
Do I need to label AI product photos on Shopify?
Yes, when the photo looks real and shows the product differently or better than it is, or shows a realistic AI model. A real product on an AI background, or a normal edit, needs no label.
Is the IPTC metadata in my AI images enough as a label?
No. The Commission says deployers cannot rely on the machine-readable marking to meet their disclosure duty. Keep the metadata for Google Merchant Center and add a visible label next to the image.
Do AI-written product descriptions need a label under the EU AI Act?
Generally no. The Commission guidelines place product descriptions outside the text duty, except claims about health, consumer safety or sustainability. Have a person with relevant knowledge review those claims. You stay responsible for every claim either way.
Does my Shopify chatbot need to say it is an AI?
Yes, at the first message. The duty sits with the app provider, or with you if you built the chatbot yourself. The ACM and AP also expect a way to reach a person.
Can a small store be fined under the EU AI Act?
The maximum for Article 50 is 15 million euros or 3% of worldwide turnover, and SMEs face the lower amount. In the Netherlands, fines under the AI Act have to wait until the Dutch implementing law is adopted.
The short version
For most Shopify stores, the EU AI Act comes down to two habits. The chat says it is an AI at the first message, and a visible label sits next to any AI image that looks real and could mislead about the product or the person in it. Health, safety and sustainability claims get a qualified human review. If you would like a second pair of eyes on how AI shows up across your store, you can start a conversation with us.
This article was drafted with AI assistance and fact-checked against the sources linked above before publishing.