AI Images in Ecommerce: The Complete Compliance Guide

The integration of artificial intelligence into ecommerce has structurally transformed the digital market. Today, many ecommerce sites use virtual try-ons, AI-generated people to model products, hyperrealistic settings… However, this level of realism has deteriorated the presumption of authenticity of digital content. In response, the European Union has approved Regulation (EU) 2024/1689, which turns transparency into a legal obligation for operating within the European market.

In this article we will develop a guide so that your ecommerce website complies with the obligations around image labeling.

1. The role of the ecommerce site: Provider or deployer?

Within the framework of the AI Act, the distinction between being a provider or a deployer defines the scope of your legal responsibilities.

Most ecommerce businesses typically act as responsible for deployment, meaning you use a third party's AI system under your own professional activity. This happens, for example, when:

• You use a Shopify app to generate virtual fashion models.
• You have an AI-powered customer service chatbot, such as Zendesk or Intercom.
• You generate your website's visual content with tools like Midjourney or DALL-E.

In this role, your main obligation revolves around transparency: you must make it clearly visible when content has been generated by AI, so the user is not misled.

However, the obligations are different when you are the one developing the AI system, or when a third party develops it for you but you release it to market under your own name or brand. Here the requirements increase significantly, because you must integrate technical marking mechanisms into the system itself so that the content can be identified as artificial from its origin.

2. What content do I need to identify?

According to the AI Act, providers of systems that generate synthetic audio, image, video, or text must ensure that these outputs are marked in a machine-readable format and are technically detectable. Many ecommerce businesses think this new regulation doesn't affect them because they aren't the ones generating AI images themselves. However, AI-generated content often makes its way onto a website through various channels:

• Stock photos: many stock image libraries have begun including files generated or modified with generative AI without the user being fully aware of it.
• Third-party images: files delivered by suppliers, agencies, or external designers who use AI without your knowledge.
• Virtual staging: tools that generate backgrounds, scenes, or even artificial models to present products.

It's very important to understand that the Regulation does not require that you personally created the image with AI. The origin of the image is irrelevant, what matters is that if the image has been generated or manipulated by an AI system and you publish it, you have the legal duty to disclose this. Therefore, the labeling responsibility falls on whoever publishes the content, not only on whoever generated it. For example, if you upload an image sent to you by an external supplier and it turns out to be a deepfake, the legal obligation to label it is yours, even though you didn't create it. For this reason, reasonable diligence is required to understand the origin of the catalogs you work with.

Does this mean every image must indicate it was generated by AI? No! Not all edited content requires a label, but the AI Act requires mandatory transparency when the content could cause confusion about its authenticity. Therefore, only deepfake cases need to be identified. In your ecommerce, this applies when virtual models have AI-generated faces and bodies wearing the products you want to sell, and it's not possible to distinguish whether it's a real person or not.

📌 And if you want to learn more about how the AI Act affects your ecommerce, we've published everything you need to have ready before August 2nd on LinkedIn. Don't miss it! Read it here.

3. What doesn't need to be identified?

To avoid unnecessary visual overload and focus transparency on content that could genuinely mislead, the Regulation and the European Commission's Code of Practice define several cases where labeling is not mandatory.

• Technical adjustments and standard editing support
  Labeling is not necessary if the AI performs a technical support function that does not substantially alter the meaning, semantics, or authenticity of the original content. This includes:
  • Quality improvements through resolution upscaling of a photograph
  • Minor corrections: lighting, contrast, color adjustments, red-eye removal, or horizon leveling
  • Basic text editing: grammar, spelling corrections, or formatting changes that don't alter the substance of the message
  • Technical processing: file compression, image rotation, or limited video stabilization

• Unaltered product images
  As already explained, the labeling obligation is only triggered when the content constitutes a deepfake. That is, when the image could be mistaken for a real person, object, place, or event. So not every AI-retouched image falls under this obligation: it's necessary to distinguish which part of the image is being altered.
  
  If what's AI-generated is the environment, for example, the background, the setting, or the scene, but the product is shown as it truly is, there is no labeling obligation. The product remains the real, faithful element of the image, and an AI-generated background doesn't create the risk of deception the rule aims to prevent.
  
  The situation changes if it's the product itself that is altered. If the AI modifies its color, dimensions, or appearance to make it look different from how it actually is (for example, "inflating" a backpack with AI to make it look fuller or bigger than it is, or changing the shade of a garment relative to its real color), that image could mislead about the actual characteristics of the product the customer will receive. In that case, if it constitutes a substantial alteration that makes the content confusable with a real representation of the item, it should be identified as generated or manipulated by AI.

• Fantasy or obviously fictional content
  The labeling obligation is only triggered when content could be perceived as truthful by a reasonably attentive person. Therefore, images showing physically impossible elements, such as dragons or flying people, don't require labeling, because their artificial nature is evident from the context in which they appear.
  
  It's worth noting that in the sphere of satire and parody, works where the AI is obvious don't need to be labeled, although a minimal notice of AI use is recommended, as long as it doesn't interfere with the enjoyment of the work.

4. What happens if I don't identify the content?

Ignoring image labeling doesn't just damage customer trust, it has a direct financial impact on your ecommerce business.

If you'd like to learn in detail what happens if you fail to comply with the AI Act and how it could affect your business, we recommend reading our article, "What Happens on August 2 If Your Website Doesn't Comply with the AI Act?", where we examine the penalties and the enforcement procedure carried out by the Spanish authorities.

Our solution: AI Sentinel

To meet this need, we have developed AI Sentinel. At Lawwwing, we believe that informing about the use of AI-generated content shouldn't hurt your store's sales. That's why AI Sentinel doesn't add labels directly onto your product images. Instead, it presents the information through a floating, accessible widget that respects your website's design and doesn't affect the shopping experience.

Our solution has been designed to help you comply with the AI Act without compromising your ecommerce experience. Because in an environment where trust has become an increasingly decisive factor, brands that arrive at August 2nd with this issue resolved will not only reduce the risk of sanctions but will also gain an edge over their competition.