Start for free

Updated on 15 July 2026

Labeling AI Advertising: EU AI Act for E-Commerce & Marketing 2026

As of: July 21, 2026

If you work in D2C, e-commerce, or marketing, you've probably been asking yourself the same question for months: do you need to label your AI advertising, meaning AI-generated product images, social creatives, paid ads, and UGC content, with a visible notice? With the deadline of August 2, 2026, a theoretical debate becomes a concrete legal obligation. From this date, the transparency obligations under Article 50 of the EU AI Act take effect. This article explains what actually applies to online retailers, brands, and agencies, where the gray areas between retouching and full synthesis lie, and how you can concretely minimize the risk of a cease and desist letter over AI product photos.

Key Takeaways

  • Labeling AI advertising becomes mandatory for certain content, above all deepfakes, from August 2, 2026 under Art. 50 of the EU AI Act; the deadline remains unchanged despite the Digital Omnibus package.
  • For e-commerce and marketing, you are almost always a deployer, not a provider; your obligation is the visible disclosure of deepfakes and AI content (Art. 50(4)).
  • You should visibly label fully AI-generated, photorealistic product images and realistic-looking AI UGC creatives as "AI-generated"; pure retouching generally does not require labeling.
  • Platform labels from Instagram, TikTok, or Meta do not replace your legal labeling obligation.
  • There are two risks: regulatory fines of up to 15 million euros or 3% of global annual turnover, and, more relevant in practice, a cease and desist letter under German unfair competition law (UWG) from a competitor.
  • A visible label plus machine-readable metadata and a proof register is the most robust approach. That's exactly what our free tool at ki-kennzeichnen.de delivers.

Table of Contents

  1. Provider or deployer? Why your role determines the obligation
  2. When does the obligation apply, and what does the Digital Omnibus change?
  3. Do AI product images in your shop need to be labeled?
  4. Retouching vs. generative AI: where is the line?
  5. AI ads, UGC, and social creatives: the channel doesn't change the obligation
  6. Practical scenarios and edge cases
  7. What happens if you don't label your content?
  8. Checklist: labeling correctly in five steps
  9. Tool tip: label AI advertising and secure your proof
  10. FAQ

A quick note upfront: for most marketing and e-commerce players, the decisive question isn't "Is this AI?" but the role you occupy under the law and the type of content.

Provider or deployer? Why your role determines the obligation

The EU AI Act distinguishes two roles with very different obligations. If you use Midjourney, DALL-E, Nano Banana, Google Veo, or another image or video model within a marketing team, you're generally a deployer, not a provider. Providers are the companies that develop the AI system and place it on the market.

Role Who this is Obligation under Art. 50 Relevance for you
Provider (para. 2) OpenAI, Google, Midjourney, model developers Mark synthetic image/audio/video/text outputs machine-readably Only if you offer an AI model yourself
Deployer (para. 4) Retailers, brands, agencies, creators Visibly disclose deepfakes and AI texts on matters of public interest Applies to e-commerce & marketing almost always

Specifically, this means:

  • Provider (Art. 50(2)): must technically mark synthetic image, audio, video, and text outputs so that they are machine-readable and identifiable as AI-generated. The solution must be "effective, interoperable, robust, and reliable," as far as technically feasible, which includes IPTC/XMP metadata, C2PA content credentials, and watermarks.
  • Deployer (Art. 50(4)): must visibly disclose certain content, specifically deepfakes (image, audio, video) as well as AI texts on matters of public interest. A lighter disclosure standard applies to art and satire.

For D2C, e-commerce, and marketing, the deployer perspective (para. 4) applies almost always. We explain in detail which role leads to which obligations in the article Provider or Deployer, Article 50 of the AI Act explained simply. You'll find the basic logic of the entire labeling obligation in our pillar article on the AI labeling obligation under the EU AI Act.

When does the obligation apply, and what does the Digital Omnibus change?

Article 50 of the EU AI Act applies from August 2, 2026, exactly 24 months after the Regulation entered into force on August 1, 2024. This is the key deadline by which you must have fulfilled your transparency obligations.

The Digital Omnibus package presented in May 2026 (provisional status) caused confusion, but changes little regarding the relevant date for marketing:

  • The machine-readable marking under Art. 50(2) is set to be postponed until December 2, 2026 for systems placed on the market before August 2, 2026.
  • The core deadline of August 2, 2026 remains.
  • The visible deployer labeling under para. 4, which is exactly your obligation in e-commerce, is not postponed.

For you as a retailer or brand, this means: count firmly on August 2, 2026. The postponement affects the technical provider side for existing systems, not the visible labeling of your ads and product images. See Gibson Dunn on the Omnibus agreement and our overview of AI Act deadlines and fines for details.

Do AI product images in your shop need to be labeled?

This is where the central gray area lies. The visible deployer obligation under para. 4 hinges on the term "deepfake." A deepfake within the meaning of the AI Act is, simplified, AI-generated or AI-manipulated content that closely resembles existing people, objects, places, or events in a way that could falsely appear authentic.

This leads to the practically relevant distinction of whether you need to label your AI product images:

  • Fully AI-generated, photorealistic product image (the product exists in reality, but the photo suggests a genuine studio or lifestyle shot): tends to fall into the deepfake category; labeling is warranted.
  • Obviously artificial illustration, comic, or abstract key visual: lacks the capacity to deceive; the visible obligation generally doesn't apply here.
  • Pure retouching of a real photo (color correction, background removal, removing a cable): not full synthesis, usually no deepfake labeling required, but be careful about misleading product representation (see below).

Important: alongside the AI Act, competition law (UWG) also applies to marketing. An AI image that shows the customer a product with a quality or effect that doesn't actually exist can be misleading and thus grounds for a cease and desist letter, entirely regardless of whether formal AI labeling is required. If you want to label AI images in marketing and use them correctly, you need to keep both levels in mind. We go deeper into the full framework for image assets in the guide Labeling AI images correctly; marketplace-specific details are covered in Labeling AI product images on Amazon & eBay.

Retouching vs. generative AI: where is the line?

The most common question in practice is: "At what point is it AI enough?" There's no sharply defined legal percentage threshold. As a guide, use the extent to which the result departs from a documentary photograph:

Editing Example Labeling under para. 4?
Classic retouching Exposure, background removal, dust correction Usually no
Partial AI editing AI swaps the background, adds to the scene, generative fill Gray area, label if in doubt
Fully AI-generated Complete scene image from a prompt Yes, if photorealistic/deceptive
AI avatar / synthetic testimonial Virtual creator speaking to camera Yes (deepfake)

The further an image departs from a documentary photograph and the more realistic the fiction appears, the more you should label it. For the official visual standards, see the article on EU icons for AI content; the technical implementation of marking is explained in C2PA, IPTC & watermarks overview.

AI ads, UGC, and social creatives: the channel doesn't change the obligation

A widespread misconception: "Instagram and TikTok automatically label AI content anyway, so I'm off the hook." That's wrong. Platform labels are based on the platform's terms of service, not on Art. 50 of the AI Act. They do not replace the legal deployer labeling, if only because you can't rely on the automatic detection working, and because the responsibility remains with you as the advertising company.

For paid ads, virtual testimonials, and AI UGC, the rule is: if the creative shows a realistic-looking person, scene, or product situation that is synthetic, deepfake disclosure and UWG apply in parallel. AI-generated testimonials and avatars that imitate a real person are particularly sensitive; here, the deepfake characteristic is present in practically every case. We cover how Meta, Instagram, TikTok, and Google Ads handle this in detail, and what else you need to consider, in the article Labeling AI advertising on Meta, Instagram & Google Ads. For video content, the guide Labeling AI videos correctly and the deep dive on Deepfake labeling are also worth reading.

Practical scenarios and edge cases

Theory only helps so much. Here are five typical situations from everyday e-commerce and a pragmatic classification:

  • Scenario 1, AI lifestyle image for a real piece of furniture: You generate a photorealistic living room in which your real, existing sofa is placed. The viewer takes it for a real photo. → Label it. Deepfake characteristic present; also observe UWG (color/dimensions must be accurate).
  • Scenario 2, AI model wearing your clothing item: A synthetic model that doesn't exist presents a real product. → Label it. A realistic-looking but fictional person is close to a deepfake.
  • Scenario 3, background removal plus AI-generated shadow: You cut out your product photo and let AI add a realistic shadow. → Gray area, note "partially AI-edited" if in doubt.
  • Scenario 4, abstract AI key visual in a newsletter: A recognizably artificial, illustrative header image without any deceptive character. → Usually no obligation under para. 4, though transparent practice is still recommended.
  • Scenario 5, virtual home staging of a property: AI furnishes an empty apartment photorealistically. → Label it and clarify that the furnishing isn't actually present; this also prevents misleading prospective buyers.

The basic rule across all cases: the more realistic the fiction and the closer it is to a purchase decision, the clearer the labeling should be.

What happens if you don't label your content?

Two levels of risk need to be distinguished:

  1. Regulatory sanctions (AI Act): Violations of the transparency obligations under Art. 50 can be punished with fines of up to 15 million euros or 3% of global annual turnover. The circulating "35 million fine" does not apply to labeling violations; that maximum rate applies exclusively to prohibited practices under Art. 5.
  2. Cease and desist letters (UWG): More relevant in practice for small and medium businesses is competition law. Competitors and industry associations can send cease and desist letters over misleading or undisclosed AI content. A cease and desist letter over AI product photos doesn't land on the EU authority's desk, it lands in your mailbox, potentially shortly after the deadline, complete with a declaration to cease and desist and a cost notice.

You can read more about deadlines, fine amounts, and the 15 vs. 35 million myth in the article AI Act deadlines and fines; how a cease and desist procedure actually unfolds is shown in AI labeling: fines and cease and desist letters.

Checklist: labeling correctly in five steps

  1. Inventory: Record which assets (product images, ads, UGC, home staging, videos) are AI-generated or AI-edited.
  2. Classify: Fully synthetic and photorealistic → label. Obviously artificial → review. Pure retouching → document.
  3. Label visibly: Place a clear notice on the content itself, not hidden in the imprint or footer. Something like "AI-generated" is acceptable. Vague phrasing such as "stock image" or "digital art" that doesn't point to the AI origin is not sufficient.
  4. Add machine-readable data: Where you're also subject to provider-adjacent obligations, or want extra assurance, write IPTC/XMP metadata or C2PA content credentials into the asset.
  5. Keep proof: Maintain a register documenting which asset was labeled when and how, your best protection in case of a cease and desist letter or an audit.

The official EU icons for AI labeling (sets for "fully AI-generated" and "partially AI-modified," each as SVG and PNG) are optional but recommended; they create a uniform, Europe-wide understandable visual standard. Our AI labeling checklist offers a complete implementation guide; you can look up unclear terms in the glossary on AI labeling.

Tool tip: label AI advertising and secure your proof

Our free web tool at ki-kennzeichnen.de covers exactly these five steps: you drag and drop the official EU icons onto images and videos, optionally add the text "AI-generated," write machine-readable IPTC/XMP metadata (C2PA plus an invisible watermark on request), and store every labeling action in the register with SHA-256 hash, timestamp, and CSV export. Batch processing (up to 50 files as a ZIP), a "check file" function, and a REST API make labeling scalable for everyday shop and campaign work, local, GDPR-friendly, and without your creatives ever leaving a third-party server.

FAQ: AI labeling in e-commerce & marketing

Do I need to label AI-generated product images in my shop?

If a fully AI-generated image looks photorealistic and depicts a real product or a real scene in a deceptively convincing way, it falls under the deepfake disclosure requirement of Art. 50(4) and should be visibly labeled as "AI-generated." Additionally, under competition law, the image must not mislead consumers about the actual nature of the product.

From when do I need to label AI advertising?

From August 2, 2026. From this date, the transparency obligations under Art. 50 of the EU AI Act apply. The Digital Omnibus only postpones the machine-readable provider marking for existing systems (placed on the market before August 2, 2026) to December 2, 2026; your visible deployer labeling remains due on August 2, 2026.

Do I need to label AI product images on Amazon or eBay?

Yes, the legal obligation applies across marketplaces and regardless of channel. You as the retailer remain responsible, even if the marketplace has its own AI policies. Don't rely on platform-side automatic labels; label the asset yourself.

What notice is sufficient for labeling AI product photos?

The notice must clearly point to the AI origin and be visible on the content itself. Phrasing like "AI-generated" or "created with AI" is suitable. Vague statements such as "stock image," "digital art," or a hidden note in the imprint are not sufficient.

Who is responsible, the agency or me as the brand?

Generally, whoever deploys the content as a deployer, meaning whoever publishes it, is responsible. If you hire an agency, the advertising brand remains obligated toward customers and competitors. So regulate labeling contractually and have proof delivered to you.

Do I need to label images retouched with Photoshop or AI?

Pure retouching of a real photo, such as exposure correction, background removal, or removing minor distractions, generally doesn't trigger deepfake labeling. However, as soon as AI generates significant portions of the image or recreates the scene photorealistically, you should label it if in doubt. Misleading representations remain relevant under UWG regardless.

Is the platform label from Instagram or TikTok sufficient for Art. 50?

No. Platform labels are based on the providers' terms of service, not on the AI Act. They don't replace your legal deployer labeling, especially since automatic detection isn't reliable. The responsibility for correct disclosure remains with you.

Do I need to label virtual home staging or furnished real estate images?

AI-furnished or virtually staged real estate photos deliberately look like genuine shots and can therefore fall under deepfake disclosure. Visibly label such images as AI-edited and clarify that the furnishing isn't actually present; this also prevents misleading prospective buyers.

What happens if I don't label AI advertising?

Regulatory fines of up to 15 million euros or 3% of global annual turnover are possible. More relevant in practice is a competition-law cease and desist letter from competitors or associations, which can lead to a declaration to cease and desist and cost reimbursement.

Do I need to label AI-generated ad copy?

For pure product texts or ad copy, there's generally no visible labeling obligation. The deployer text obligation under para. 4 targets AI texts on matters of public interest. Details can be found under Labeling AI texts & ChatGPT content.

Sources

This article reflects the status as of July 21, 2026, and does not constitute legal advice. For your specific case, please consult legal counsel.

Read more