Labeling AI Product Images on Amazon: What Becomes Mandatory in 2026

If you want to label AI product images on Amazon or list them on eBay, Otto, Kaufland, and Etsy, from 2026 you face two sets of rules at once: the platform's policies and the EU AI Act. Both demand transparency for certain AI content, above all deceptively realistic images, but they apply at different points and not equally to every AI photo. The central deadline is August 2, 2026: from this date, the transparency obligations under Article 50 of the EU AI Act apply. This article breaks down the legal situation marketplace by marketplace, explains the practical metadata problem on upload, and shows you, as a seller, how to cleanly rule out cease-and-desist risk, even with hundreds of SKUs. Last updated: July 21, 2026.

Key Takeaways

• From August 2, 2026, Article 50 of the EU AI Act applies. As an Amazon, eBay, Otto, Kaufland, or Etsy seller, you are almost always a deployer: AI product images that deceptively fake a real scene (deepfakes) must be visibly disclosed; a neutral product photo often does not, so check the individual case.
• Two sets of rules apply in parallel: the platform's terms of service (contract law) and the law (public law). Follow the stricter standard.
• Metadata alone is not enough: marketplaces recompress and reconvert images on upload, and IPTC/XMP, C2PA manifests, and invisible watermarks are frequently lost in the process.
• You need two things: a visible label in the image plus a notice in the listing, and your own evidence (a register with SHA-256 checksums).
• The real risk for sellers is less the regulatory fine (up to 15 million euros / 3% of turnover) than the fast, costly competition law cease-and-desist notice.
• The Digital Omnibus only postpones the machine-readable marking for legacy systems to December 2, 2026; the visible deployer obligation (paragraph 4) stays at August 2, 2026.

Contents

1. Two sets of rules: platform policies and law
2. What the EU AI Act requires from August 2, 2026
3. Marketplace by marketplace: Amazon, eBay, Otto, Kaufland, Etsy
4. The metadata problem on upload
5. Cease-and-desist risk for sellers: where the danger is real
6. Edge cases: when you must label, and when not
7. Step by step: label your entire catalog
8. Checklist for marketplace sellers
9. FAQ

Two Sets of Rules: Platform Policies and Law

For marketplace sellers, the key insight is: platform policies and the law exist side by side and do not replace each other. Meeting a platform's terms of service alone does not automatically make you legally compliant, and vice versa.

Platform rules are contract law. Amazon, eBay, and others may set their own requirements for product images in their terms of use, for example that the main image must show the real product against a white background and must not be misleading. Violations don't lead to a regulatory fine but to consequences from the platform: listing suppression, loss of visibility, and in extreme cases account suspension. These rules change continuously and vary in strictness by marketplace.

The law, specifically Article 50 of the EU AI Act, is a public law obligation. It applies regardless of what a platform's terms of service say, and it applies to you as the business deploying the content. Even if a marketplace would theoretically tolerate AI images, that does not release you from the statutory transparency obligations.

The practical consequence: follow the stricter of the two standards. Anyone who meets the statutory requirements and also observes the platform policies reduces the risk the most. Our guide on AI labeling in e-commerce & marketing explains the basic framework; if you want to know whether and how strongly the obligation applies to you at all, you will find the overview under AI labeling obligations under the EU AI Act.

What the EU AI Act Requires From August 2, 2026

Article 50 of the EU AI Act distinguishes two roles, and this distinction determines your specific obligations as a seller.

Provider, machine-readable marking (Art. 50(2)): whoever provides an AI system that generates synthetic image, audio, video, or text content must mark these outputs as AI generated in a machine-readable way, for example via IPTC/XMP metadata, the C2PA standard, or watermarks. According to the regulation, the marking must be "effective, interoperable, robust, and reliable, to the extent technically feasible." This obligation primarily applies to the operators of the image generators (e.g., Midjourney, OpenAI, Adobe Firefly), not to the seller.

Deployer, visible disclosure (Art. 50(4)): whoever uses AI content must visibly disclose deepfakes; the same applies to AI texts on matters of public interest. As a D2C brand, Amazon seller, or eBay merchant, you are generally a deployer. A realistic-looking product photo that deceptively fakes a real scene can fall under the deepfake concept of the transparency obligation. A neutral, clearly promotional product photo often does not, so a case-by-case check is worthwhile. A lighter standard applies to art, satire, or clearly fictional depictions. The article Provider vs. deployer: who has which obligations covers the exact distinction between the roles.

One detail from the so-called Digital Omnibus (provisional, as of May 2026): the machine-readable marking obligation (paragraph 2) for AI systems placed on the market before August 2, 2026 has been postponed to December 2, 2026. The core deadline of August 2, 2026 remains in place, however, and the visible deployer labeling (paragraph 4) was expressly not postponed. For you as a seller, the postponement therefore changes nothing about the visible labeling obligation from August 2, 2026.

The official EU icons (available as SVG and PNG, including the variants "fully AI-generated" and "partially AI-modified") are an optional offering from the Commission, not mandatory, but practical because they are understandable across language boundaries. That pays off especially in cross-border marketplace sales. Violations of the AI Act carry a risk of sanctions of up to 15 million euros or 3% of global annual turnover, whichever amount is higher.

Marketplace by Marketplace: Amazon, eBay, Otto, Kaufland, Etsy

The statutory obligation is uniform EU-wide, but the platform environment in which you implement it is not. Here are the practically relevant differences.

Amazon. Amazon sets high requirements of its own for product images: the main image must show the real product, no misleading additions, a defined background. It's important not to confuse this image policy with the statutory obligation: even a technically "clean" Amazon main image must be labeled as soon as it is AI generated. Because Amazon aggressively rescales images on upload and converts them into its own formats, a visible in-image label is particularly important here; metadata often does not survive the process.

eBay. eBay is more liberal in its image requirements and allows more creative freedom. That is exactly what increases the likelihood that sellers deploy AI lifestyle images without labeling. The statutory obligation applies identically on eBay; the more liberal terms-of-service framework is not a free pass.

Otto Market and Kaufland.de. Both are open marketplaces with connected third-party sellers, who are mostly supplied via middleware and PIM systems (e.g., via CSV or API feed). Images here usually run automatically through a processing pipeline, another point where metadata can get lost. Anyone distributing via a feed should therefore handle labeling at the source file (visible label embedded) before the feed reaches the platform.

Etsy. Etsy lives on the "handmade" and authenticity promise. AI-generated product images are in particular tension with buyer expectations here; the reputational and misleading-conduct risks are elevated in addition to the AI Act obligation. Label more generously here.

The same pattern applies across all platforms: different terms of service, identical legal situation. The common denominator is a visible label plus documented evidence, regardless of how the respective platform subsequently processes your file.

The Metadata Problem: Why You Need a Visible Label

This is where the real crux lies for marketplace sellers. Many rely on the assumption that their AI tool marks the images in the metadata anyway. In practice, that is not enough on marketplaces.

Marketplaces strip metadata. When you upload an image into a listing, the platform processes the file: it rescales, recompresses, and reconverts the image, and generates its own preview sizes and WebP variants. During this process, IPTC/XMP fields, C2PA manifests, and sometimes even invisible watermarks are frequently lost. The image your customers ultimately see then no longer carries the machine-readable marking.

A concrete example: you generate a lifestyle photo of your water bottle with a C2PA manifest and the IPTC field "DigitalSourceType: trainedAlgorithmicMedia." On Amazon upload, the 4000 pixel file is downsampled to several target sizes and served as a fresh JPEG/WebP, without your metadata. If an authority or a competitor later checks the publicly visible image, no AI marking can be found anymore. Your cleanly marked original then only helps you if you have kept and documented it separately.

Two practical consequences follow from this:

1. A visible label is indispensable. Since you cannot rely on the metadata surviving, the AI labeling should be embedded in the image as a visible element, for example a subtle EU icon or a text addition in the corner of the image, supplemented by a clear note in the item description. This simultaneously fulfills the visible disclosure obligation for deployers and survives any recompression.

2. You need your own evidence. Because the platform alters the file, you should document independently which images you labeled how and when. A compliance register with checksums (e.g., SHA-256) proves, in a dispute, that you fulfilled your obligations, even when the uploaded file no longer carries any metadata later on.

So combine both: the machine-readable marking on the source file (for evidence and standard compliance) and a visible label plus a written notice in the listing. We explain the technical foundations of machine-readable marking in the article Labeling AI images.

Cease-and-Desist Risk for Sellers: Where the Danger Is Real

For sellers, the immediate risk is not primarily the regulatory fine but the competition law cease-and-desist notice. Missing or incorrectly labeled AI product photos can be assessed as misleading business conduct, for example when an AI image shows a product more attractively, in nonexistent color variants, or in a usage situation that does not actually occur.

Cease-and-desist notices typically come from competitors or associations entitled to issue them, and they are fast and costly. This pattern is familiar from other e-commerce topics (right of withdrawal notices, unit price disclosure, data protection): as soon as a new obligation takes effect, heightened cease-and-desist attention typically follows. Around August 2, 2026, sellers should therefore take the topic of "cease-and-desist notices for AI product photos" seriously. You can read about what threatens in a worst case and how to respond under AI labeling: penalties & cease-and-desist notices.

It becomes especially delicate when the AI image deviates from the product actually delivered. Then, alongside the transparency obligation, classic misleading-conduct and warranty questions also come into play. Returns received due to "image deviates from product" are also a tangible economic problem. Basic rule: AI may stage, but not deceive, and the AI origin must remain recognizable.

Edge Cases: When You Must Label, and When Not

Not every digital edit triggers the labeling obligation. What matters is whether generative AI substantively creates or materially alters the image content. The following overview classifies typical marketplace scenarios:

The AI Act expressly distinguishes between "fully AI-generated" and "partially AI-modified," which is why there are two separate EU icons for this. Rule of thumb for practice: as soon as a viewer could gain a different impression of the product through the AI editing, label it. For purely technical optimization without inventing content, no labeling is required.

Step by Step: Label Your Entire Catalog

If you only maintain a handful of listings, you can label images one by one. But as soon as you distribute dozens or hundreds of product images across multiple marketplaces, you need a scalable, provable process. Here is how you proceed:

1. Review your inventory. Record which listing images were fully or partially created with AI. If unsure, check the source file for existing AI markings (IPTC/C2PA).
2. Classify edge cases. Assign each image using the table above: fully AI-generated, partially AI-modified, or exempt from labeling.
3. Embed a visible label. Give images requiring labeling the matching EU icon in a fixed corner of the image, so that it remains recognizable after Amazon/eBay compression.
4. Mark in a machine-readable way. In parallel, write IPTC/XMP, C2PA, and a watermark into the source file, for channels that retain metadata and for your own evidence.
5. Add a notice to the listing. Add a clear sentence to the item description, e.g. "Product image partially created with AI."
6. Document evidence. Log every file with a SHA-256 checksum in a compliance register and export it (CSV) for your records.
7. Automate the process. Connect the workflow to your upload or feed process via API or batch processing, so new images run through it automatically.

With many SKUs, step 7 is decisive. An assortment of 300 items with 5 images each means 1,500 files, which cannot be sensibly labeled and documented by hand. This is exactly what our free tool AI labeling at ki-kennzeichnen.de is built for. It covers both levels you need on marketplaces:

• Visible EU label via drag and drop, the official EU icons are embedded directly into images and videos, so the labeling survives the metadata stripping on Amazon, eBay, and others.
• Machine-readable marking via IPTC/XMP, C2PA, and watermarks for the source files and channels that retain metadata.
• Check file: check existing images for existing AI markings.
• Compliance register with evidence (SHA-256 checksums, CSV export), your independent proof that every image was labeled properly.
• Batch processing: up to 50 files at once, output as a ZIP, ideal for entire product catalogs.
• REST API to connect to your PIM or your feed/upload workflow (e.g., for Otto and Kaufland feeds).

The difference from purely client-side solutions: because processing runs on a real backend, you get a register, verifiable evidence, batch capability, and an API, the four functions that make the difference, in everyday marketplace work, between "labeled somehow" and "verifiably documented." You can use the tool here for free.

Checklist for Marketplace Sellers

• Identified all AI-generated and AI-modified listing images
• Classified each image as fully / partially AI or exempt from labeling
• Embedded a visible EU icon in images requiring labeling
• Marked the source file in a machine-readable way (IPTC/XMP, C2PA, watermark)
• Added a notice text to the item description
• Set up a compliance register with SHA-256 and CSV export
• Additionally checked the platform policies of Amazon, eBay, Otto, Kaufland, Etsy
• Automated the upload/feed process for new images (batch/API)

Anyone who additionally runs paid campaigns for their listings should observe the rules under AI advertising on Meta & Instagram; the same transparency principles apply there to advertisements.

FAQ: Labeling AI Product Images on Amazon and eBay

Do I even need to label AI product images on Amazon and eBay? From August 2, 2026, Article 50 of the EU AI Act requires transparency for AI-generated content. As a seller, you are generally a deployer and must visibly disclose AI images when they qualify as deepfakes, that is, look deceptively real; a neutral product photo often does not. The respective platform policies also apply. Follow the stricter standard.

From exactly when does the obligation apply? From August 2, 2026. That is the core deadline for Article 50, 24 months after the EU AI Act entered into force on August 1, 2024. There is no postponement for the visible labeling you owe as a deployer.

Is it enough to mark the image's metadata? No. Amazon, eBay, Otto, and Kaufland reprocess uploaded images and frequently remove IPTC/XMP, C2PA, and watermark information in the process. So rely on a visible label in the image plus a notice in the description, and document the labeling separately in a register.

Do the same rules apply on Otto, Kaufland, and Etsy as on Amazon? The statutory obligation under the EU AI Act is identical EU-wide and platform-independent. Only the respective platform terms of service on image design differ. On Etsy, an elevated risk of misleading conduct is added because of the "handmade" promise.

Who is liable, the AI provider or me as the seller? Both, on their respective level. The AI provider is responsible for the machine-readable marking (paragraph 2). You, as the deployer, are responsible for the visible disclosure in the listing (paragraph 4). You cannot shift this obligation onto the platform or the tool.

What happens to me if labeling is missing? Two levels: under the AI Act, sanctions of up to 15 million euros or 3% of global annual turnover are possible. More practically relevant for most sellers is the competition law cease-and-desist notice from competitors or associations, which arrives quickly and comes with costs.

Does the obligation also apply to photos only lightly edited with AI? The AI Act distinguishes between fully AI-generated and partially AI-modified content; there are even dedicated EU icons for this. As soon as an AI edit materially changes the impression of the product, you should label it. Plain standard retouching (cropping, brightness) generally does not fall under this.

Do I have to use the official EU icons? No, the EU icons are an optional offering from the Commission. But they are recommended because they are understandable regardless of language and cleanly represent the "partially" and "fully AI-generated" cases. A clearly legible text addition is also acceptable.

How do I efficiently label an entire product catalog? Via batch processing: with the AI labeling tool, you upload up to 50 images at once, give them a visible EU label and machine-readable marking, and get a CSV register as evidence. Via the REST API, this can be integrated into existing PIM and feed processes (Otto, Kaufland).

Was the labeling obligation postponed by the Digital Omnibus? Only partially. The machine-readable marking for legacy systems (placed on the market before August 2, 2026) has been postponed to December 2, 2026. The core deadline of August 2, 2026 and the visible deployer labeling remain unchanged.

Sources

• European Commission, timeline for implementing the EU AI Act: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
• Article 50 EU AI Act (full text): https://artificialintelligenceact.eu/article/50/
• European Commission, EU icons for labeling AI-generated content: https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content
• European Commission, guidelines on transparency obligations for providers and deployers: https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems

This article does not replace legal advice. For your specific case, please consult qualified legal advice. Last updated: July 21, 2026.