Skip to content
Shopware & E-Commerce

AI product images on Amazon, eBay, Etsy, Instagram and TikTok: what each platform requires for labelling – and what applies in your own store

The same AI-generated product image ends up today in your own store, in the Amazon listing, on eBay, in the Google Shopping feed and as an Instagram post. Five channels, five rulebooks – and since August 2, 2026 one law that sits above all of them. Anyone who generates or edits product images with AI needs to know where a visible label is mandatory, where it's banned and where the platform sets it itself as soon as it finds metadata.

On November 7, 2026 we read the policies of twelve platforms – Seller Central, partner universities, community guidelines, ads policies – and checked them against the AI Act and the Commission's guidelines. The result is less contradictory than it looks once you keep one distinction in mind: visible label versus metadata in the file.

The short version

  • No marketplace bans AI product images. All of them require that the image shows the product exactly. Misleading AI images fail on that rule, not on an AI ban.
  • A real labelling duty exists on Google Merchant Center (IPTC metadata in the file), Etsy (disclosure in the description plus at least one real photo since October 22, 2026), TikTok and YouTube (label realistic AI content). Amazon has required an XMP keyword for photorealistic AI people only since July 23, 2026. eBay, Pinterest and LinkedIn require nothing.
  • Visible labels collide with image rules: Amazon main image without text, eBay without text and watermarks, Zalando without overlays, Google Shopping without overlays. Zalando instead requires an invisible mark in the metadata.
  • Art. 50 AI Act applies on every channel to you as the deployer – in the store, in the listing, in the post. According to the guidelines the platform isn't the deployer as long as it only distributes your content. Our plugin EU AI Act Compliance implements the duty in your Shopware store; for the marketplaces, metadata remains your tool.

As always: we're software engineers, not lawyers. Quotes come from platform policies, the regulation and the guidelines; the framing is our practical summary – not legal advice. Platform rules change often; every statement carries its date.

What you have to do per channel

  • Your own Shopware store

    Visible badge on deepfake-capable images, read metadata

    Art. 50(4) AI Act requires visible disclosure at the latest at first exposure – in listing, gallery, zoom and shopping experiences. The plugin sets the badge automatically and reads C2PA and IPTC data on upload as a suggestion.

  • Google Shopping and Merchant Center

    IPTC metadata in the file, no overlay

    Google requires the IPTC value “TrainedAlgorithmicMedia” in the file for every AI image and forbids removing it; overlays and watermarks in the image are prohibited. AI text goes into the “structured_description” attribute.

  • Amazon, eBay, Zalando

    Real main image, keep metadata, Amazon’s XMP keyword

    Main images may not carry text, Zalando bans visible labels outright. Since July 2026 Amazon requires the keyword “contains-synthetic-performer” in the dc:subject field for photorealistic AI people; Zalando wants an invisible mark by December 2026.

  • Instagram, TikTok, YouTube, Pinterest

    Label it yourself where realistic – metadata does the rest

    TikTok and YouTube require a label for realistic-looking AI content, Meta only for video and audio. All four read C2PA or IPTC and set their own label automatically when the file still carries the data.

Twelve platforms compared

"Not verified" means: we found no public policy – for Otto Market and Kaufland the partner rules sit behind a login.

AI product images on marketplaces, social media and Google (as of November 7, 2026, per platform policies)
PlatformAI images allowedLabel mandatoryfor sellersAutomatic labelby the platformReads C2PA / IPTCSanctionper policy
Amazon (listings, A+)yesimage must show the product “as a realistic image”partlyphotorealistic AI people only: XMP “contains-synthetic-performer”, since Jul 23, 2026partlyindicator “where applicable”, sellers report not visiblepartlyown XMP keyword; C2PA not verifiedsuppression for image violations
eBayyesmust “accurately represent” the item; no stock photos for used goodsnopartlyown AI model images only (test 04/2026)not verifiedlisting removal
Etsypartlysupplementary only; at least one unaltered real photo since Oct 22, 2026yesdisclose AI use in the descriptionnonot verifiedremoval, refunds to buyers
Zalando Partneryesrejected for “AI distortions”partlyno visible label allowed; invisible mark by 12/2026nopartlyrequires metadata, reading not verifiedimage rejection
Otto Market, Kauflandnot verifiednot verifiednot verifiednot verifiednot verified
Meta – Instagram, Facebook organicyespartlyphotorealistic video and audio; AI personas since Aug 31, 2026yes“AI info” via industry signalsyes“may apply penalties”, reach limit
Meta Adsyespartlypolitics and social issues onlyyesMeta AI tools and detected third-party signalsyes–
TikTok organicyesyesrealistic-looking scenes and peopleyesTikTok effects, C2PA, own detectionyesC2PA since 05/2024removal, restriction, forced label
TikTok AdsyesyesAIGC label or own disclaimerpartlypartlyper secondary sourcesad rejected or restricted
YouTubeyesyesrealistic-looking AI or altered contentyesYouTube AI tools, C2PA, internal detectionyesremoval, suspension from the Partner Program
Pinterestyesnoyes“AI modified” via IPTC and classifieryesIPTCnone
LinkedInyesnoyesC2PA icon on signed contentyesC2PAnone
Google Merchant Center, ShoppingyesyesIPTC “TrainedAlgorithmicMedia” in the file; AI text in structured_descriptionnoyesIPTCrejection for misrepresentation
Google Adsyespartlyregional: EU, India, New YorkyesGoogle AI tools automatically, “How this ad was made”yesSynthID and C2PAGoogle label cannot be removed
Google Search, Imagesyesnopartly“About this image” shows C2PA and IPTC infoyesIPTC, C2PA from 2.1 with trust listspam measures for scaled content

Legend:yespartlynonot verified

Sources: Amazon Seller Central (guide G1881, news of July 23, 2026), eBay picture standards and user agreement (June 28, 2026), Etsy creativity standards (June 10, 2025) and news (October 22, 2026), Zalando Partner University (October 1, 2026), Meta Newsroom (2024, February 3, 2025), Instagram (August 31, 2026), TikTok Community Guidelines (effective September 24, 2026) and ads policy (April 2026), YouTube help, Pinterest help (April 30, 2025), Google Merchant Center and Google Ads (July 2026), LinkedIn help. “Not verified” = no public policy found.

The law applies on every channel

Before the platforms, the foundation: Art. 50(4) AI Act obliges deployers of an AI system that generates deepfakes to disclose visibly. A deployer is anyone using the system "under their authority" – the merchant who generates a product image with Midjourney or has an agency generate it. The duty attaches to the deployer, not to the medium. It applies identically in your own store, in the Amazon listing and in the Instagram post.

The platform itself isn't the deployer. According to a law firm's summary, the final Commission guidelines make clear that hosting services and online platforms aren't deployers "where their role is limited to disseminating AI-generated content created by third parties". They become deployers only for their own AI images – for instance when eBay inserts AI models into listings or Otto generates model photos with its own system.

So why do Meta, TikTok, YouTube and Pinterest label at all? Because the Digital Services Act requires it of very large platforms: Art. 35(1)(k) DSA lists, as a risk-mitigation measure, that generated or manipulated content that "falsely appears to a person to be authentic or truthful" must be "distinguishable through prominent markings". That explains the automatic labels – and why they only kick in when your file still carries the signals.

What a deepfake is, Art. 3(60) defines: AI content "that resembles existing persons, objects, places, entities or events" and would falsely appear authentic. The guidelines cite as an example an AI product image that makes the product look more attractive or of better quality. A real product on an AI background for aesthetic purposes isn't one; cut-outs and colour correction are standard editing. How the line is drawn in detail is in our fundamentals article; the duties per content type we broke down in a dedicated matrix.

Marketplaces: Amazon, eBay, Etsy, Zalando

Amazon. The product image guide in Seller Central requires that the image shows the product "as a realistic image of professional quality" and correctly represents "the actual size, quantity and colour of the product". There is no AI ban; a misleading AI image fails on the accuracy rule. Since July 23, 2026 a worldwide disclosure duty applies to photorealistic AI-generated people: the XMP keyword contains-synthetic-performer in the dc:subject field, upon which Amazon adds "an indicator to inform customers … where applicable". The trigger was a New York law on synthetic performers. In the seller forum, merchants report seeing no visible indicator; whether it appears is open. Main images still may contain no text or logos – a visible AI label in the main image would be a violation. Beware of blogs claiming a "ban on AI faces in main images from August 2026" or six-figure suppression numbers: there is no Amazon source for either.

eBay. The picture standards require that photos "accurately represent" the item; prohibited are stock images for used or defective items as well as "watermarks of any kind" and added text in the image. There is no AI labelling duty. The US user agreement effective June 28, 2026 makes the seller expressly responsible for content created with "generative artificial intelligence tools"; the German terms contain no AI clause. The reverse case is notable: in April 2026 eBay tested AI-generated model images in listings – with an AI disclaimer in the image, set by eBay itself.

Etsy. The strictest marketplace. The creativity standards (last updated June 10, 2025) require sellers to state in the item description whether an item was made using AI. Since October 22, 2026 every listing needs at least one unaltered photo of the actual item against a real background; supplementary AI images must accurately represent the item, and bot-generated listings without human involvement are banned. Etsy announced it would expand removals and refunds buyers when a listing doesn't meet the image requirements.

Zalando. The partner image guidelines (as of October 1, 2026) allow AI images but reject "AI distortions" – heels merging into furniture, altered logos, airbrushed skin. Two rules decide the labelling question: "We don't allow visible on-image labels like text overlays" and "You must embed identifying data within the metadata or as an invisible watermark", required by December 2026. Content that would need a visible label under the law shouldn't be delivered at all. Zalando resolves the conflict between Art. 50 and clean product images by excluding deepfake-capable images.

Otto Market and Kaufland. We found no public policies on AI images; the partner documentation sits behind a login. Otto itself has been generating model photos with its own AI system since 2025 and, according to heise, saves up to 60 percent of photo costs – how the group labels these images remained open in the coverage.

Social media and video: Meta, TikTok, YouTube, Pinterest

Meta. Since 2024 Meta labels images with "AI info" when it finds "industry standard indicators" – C2PA or IPTC signals – or when you declare it yourself. For content that was only AI-edited, the label moved into the menu in July 2024; fully generated content keeps it visible. A self-labelling duty applies only to photorealistic video and realistic-sounding audio; Meta "may apply penalties" for violations. Since August 31, 2026 AI personas on Instagram must be labelled "AI-generated profile", otherwise reach is reduced. In ads, Meta has automatically labelled since February 2025 whatever was created or "significantly edited" with its own AI tools and detects third-party tools via signals; a mandatory disclosure exists only for political and social-issue ads.

TikTok. The community guidelines (effective September 24, 2026) require creators to label "AI-generated or significantly edited content that shows realistic-looking scenes or people"; unlabelled content may be removed, restricted or labelled by TikTok. TikTok has read C2PA Content Credentials since May 2024 and attaches them to downloads. For ads, since April 2026: fully AI-generated or "significantly edited" content needs the AIGC label or an own disclaimer, otherwise the ad is rejected. Important for merchants: lighting adjustments, denoising and "removing or modifying backgrounds" count as insignificant – a real product photo on an AI background isn't a label case in TikTok Ads. For TikTok Shop, secondary sources report a dedicated AI policy since May 2026 and a ban on AI voices and still images in lives; we couldn't read the original text.

YouTube. "We require creators to disclose AI-generated or altered content that seems realistic." Mandatory cases are real people saying or doing something that didn't happen, and a "realistic scene that didn't actually occur" – an AI lifestyle clip with your product probably qualifies, a colour correction doesn't. YouTube labels automatically for its own AI tools and for C2PA metadata; penalties go up to suspension from the Partner Program.

Pinterest and LinkedIn. Neither requires anything, but both label themselves: Pinterest sets "AI modified" based on IPTC metadata and its own classifiers, LinkedIn shows the C2PA icon on signed content. Pinterest users report false labels on hand-drawn art – a risk also for real product photos retouched with AI.

Google: Merchant Center requires metadata, Ads labels

Google Shopping is the channel with the clearest duty – and it's invisible. The Merchant Center help says: "All images created using generative AI must contain metadata indicating that the image was AI-generated", specifically the IPTC value DigitalSourceType = TrainedAlgorithmicMedia, for main image, additional images and lifestyle images. Embedded tags must not be removed, not even from Google's own Product Studio. At the same time Google forbids overlays and watermarks in Shopping images – the visible label has no place there. AI-generated titles and descriptions belong in structured_title and structured_description with digital_source_type.

Google Ads has gone the other way since July 2026: a panel "How this ad was made" shows whether an ad was created or edited with AI; Google's own AI tools trigger it automatically, third-party AI is labelled by the advertiser via the "AI label" setting. It's mandatory regionally – Google names the EU, India and New York – and once set, Google labels can't be removed. In Search there is no labelling duty; "About this image" shows C2PA and IPTC information when the file carries it.

The common denominator: metadata in the file

Lay the table over the image rules and a pattern remains: where a visible label is banned or unwanted – Amazon main image, eBay, Zalando, Google Shopping – the same platforms require or read metadata. Where platforms label themselves – Meta, TikTok, YouTube, Pinterest, LinkedIn, Google Search – they do so only when the file still carries C2PA or IPTC data. Metadata is the one channel that works everywhere.

The problem: it's easily lost. Screenshots, format changes, many upload pipelines and most social platforms strip it – which generators set it at all and what survives the trip is in our article on C2PA and IPTC metadata. For the multichannel routine that means:

The common denominator: metadata in the file
PlaceRule
Original fileKeep it straight from the generator, with manifest and IPTC field; never use the export from an intermediate tool as the master
Marketplace uploadUpload the original, add Amazon's XMP keyword for AI people, check Zalando's invisible mark
Shopping feedPoint image_link to files with the IPTC value; AI text in structured_description
Main imageA real product photo – mandatory on Etsy, the safe choice on Amazon and eBay, the expectation on Zalando
Social mediaLabel realistic AI content yourself (TikTok, YouTube, Meta for video), post the file with metadata so the platform label kicks in
Your own storeVisible badge under Art. 50(4), read metadata on upload, document the status

A side finding from case law that affects merchants with AI-retouched photos: according to a law firm's report, the Regional Court of Frankfurt ruled on May 27, 2026 (2-06 O 347/25) that heavily AI-edited product photos can lose their copyright-like protection as simple photographs – a competitor was allowed to copy them. Anyone using AI should know that the result may no longer be theirs.

What happened in 2025 and 2026

What happened in 2025 and 2026
DatePlatformEvent
2024AmazonAmazon replaces second listing images with AI versions on its own; sellers can't remove them
Nov 17, 2025TikTok ShopSafety report: "hundreds of thousands of fake sellers and bogus listings" removed, AI images as a driver
Apr 14, 2026eBayAI-generated model images inserted into listings without consent, partly with distorted details; eBay puts an AI notice in the image
May 23, 2026TikTok ShopBan on AI voices and still images in lives and shoppable videos, sanctions up to account bans (secondary sources)
Jul 23, 2026AmazonMandatory keyword "contains-synthetic-performer" worldwide
Aug 31, 2026InstagramReach limit for unlabelled AI personas
Oct 22, 2026EtsyMandatory real photo per listing, ban on bot-generated listings, expansion of removals

We found no documented suspension of a seller because of missing AI labelling on Amazon or eBay, nor a German court ruling or warning letter over unlabelled AI product images. Enforcement so far runs through image policies and platform interventions, not the AI Act.

What customers think of it is shown by two 2026 surveys: in a UK poll by the imaging vendor Photoroom (2,000 respondents), 55 percent said poorly executed or heavily edited AI product images made them trust a marketplace less; only 33 percent were comfortable with AI product images even when clearly labelled. In Germany, according to Verve Group and Censuswide (August 2026), 35 percent lose brand trust when advertising is recognisably AI-made, 45 percent among 18- to 24-year-olds. The lesson isn't "no AI" but: the image has to be right, and the main image stays real.

What this means for your Shopware store

In your own store there is no platform rule banning the visible label – here Art. 50(4) applies directly. Our plugin EU AI Act Compliance implements it: you maintain one of four AI statuses per medium, the badge appears automatically in listing, product gallery, zoom, shopping experiences and on videos, configurable per sales channel. On upload the plugin reads C2PA and IPTC metadata and suggests the status with a confidence score – the same data Google Merchant Center requires and Meta, TikTok and Pinterest use for labelling. Optionally there's machine-readable markup for search engines and the AI register as evidence of which image was labelled how. How to get thousands of images into the right status in one afternoon is in the hands-on article.

What the plugin can't do: label your Amazon uploads or write Zalando's invisible mark. For that, the original file with its metadata remains your tool – and the rule that the main image on every channel is a real photo.

Multichannel checklist

  1. Inventory: which product images are AI-generated or substantially AI-edited? Set the status in the store, keep originals with metadata.
  2. Real main image: an unaltered product photo as the first image on every channel – mandatory on Etsy, otherwise the safe choice.
  3. Google Shopping: IPTC value in every AI image file, no overlays, AI text in structured_description.
  4. Amazon: set the XMP keyword for photorealistic AI people; no text in the main image.
  5. Zalando: no visible labels, invisible mark by December 2026, don't deliver deepfake-capable images.
  6. Social: label realistic AI content on TikTok and YouTube, on Meta for video and audio; post files with metadata.
  7. Your own store: badge under Art. 50(4), metadata scan, register – and at every channel switch check whether the file still carries its data.

Frequently asked questions

Are AI-generated product images allowed on Amazon?

Yes. Amazon’s product image guide doesn’t ban AI but requires that the image shows the product exactly as a realistic image – size, quantity and colour correct. Since July 23, 2026 photorealistic AI-generated people in listing and A+ images must be tagged with the XMP keyword “contains-synthetic-performer”. Text and logos are banned in the main image, so a visible AI label there is too.

Do I have to label AI images on eBay?

eBay doesn’t require an AI label. Its picture standards require photos to accurately represent the item, ban stock images for used or defective goods and prohibit watermarks and text in the image. The US user agreement of June 2026 makes the seller expressly responsible for content from generative AI tools. The statutory duty under Art. 50(4) AI Act applies to you in the eBay listing regardless.

What does Google Shopping require for AI product images?

Every image created with generative AI must carry the IPTC value DigitalSourceType equal to TrainedAlgorithmicMedia in the file, and embedded metadata must not be removed – for main image, additional images and lifestyle images. Visible overlays and watermarks are prohibited. AI-generated titles and descriptions belong in the attributes structured_title and structured_description with digital_source_type.

Which platforms label AI images automatically?

Meta (Instagram, Facebook, Threads) with “AI info”, TikTok, YouTube, Pinterest with “AI modified” and LinkedIn with the C2PA icon – each based on C2PA or IPTC metadata in the file, partly supplemented by their own detection. Google Ads labels ads from its own AI tools automatically, Google Search shows the data under “About this image”. The precondition is always that the file still carries its metadata.

Does the AI Act also apply to marketplace listings?

Yes. Art. 50(4) AI Act obliges the deployer of the AI system – the merchant who generates the image or has it generated – regardless of channel. According to the Commission’s guidelines the platform isn’t the deployer as long as it only distributes third-party content. That Meta, TikTok or YouTube set labels follows from the Digital Services Act and doesn’t replace your duty.

Is a visible AI label on the product image allowed everywhere?

No. Amazon bans text and logos in the main image, eBay text and watermarks, Zalando visible labels altogether, Google Shopping overlays. On these channels metadata in the file is the way – mandatory at Google, required by Zalando by December 2026. In your own store there is no such restriction; there Art. 50(4) requires visible disclosure at the latest at first exposure.

Sources

Retrieved November 7, 2026:

Bottom line

The platform rules on AI product images look like twelve special cases and follow two lines nonetheless: the image has to show the product as it is – and the origin belongs in the file, because a visible label has no place on most marketplaces. Above all of it sits Art. 50 AI Act, which hits you as the deployer on every channel but requires a visible label only where you're allowed to set one: in your own store.

Take your best-selling product and follow its main image through every channel: is it a real photo? Does the file still carry its metadata? Is there a badge in the store where one belongs? Three questions, three minutes – and if one answer is missing, we're happy to talk about it in a free initial call.