AI generated product images: disclosure rules for Amazon, Google and Etsy

The short answer: if AI only cleaned up a real photo (background removal, color correction, exposure), none of the three channels asks you to label it. If AI created something in the image, each one does. Amazon wants the keyword contains-synthetic-performer in the XMP subject field of any image with photorealistic AI-generated people. Google Merchant Center wants an IPTC "digital source type" tag on AI-generated images and tells you not to strip it. Etsy wants a line in the listing description when an item is created with AI. Most of this disclosure lives in file metadata, and many resize tools delete metadata by default.

Last checked: October 10, 2026. These rules are new and still moving. Recheck each channel before a big upload.

[Image to add: Square card, white background. Three rows: "Amazon: XMP keyword", "Google: IPTC source type", "Etsy: listing description". Alt text: "AI image disclosure by channel: Amazon uses an XMP keyword, Google an IPTC tag, Etsy the listing description"]

Where the line sits: correcting versus creating

Every channel draws roughly the same line, so it helps to sort your AI steps before reading the rules.

Correcting what you shot. Cutouts, background cleanup, removing color spill, evening out white balance across a batch. Amazon staff on Seller Forums describe "simple AI retouching, like background removal, color correction, lighting adjustments" as "generally treated the same as traditional photo editing."

Creating something that wasn't there. A generated lifestyle scene behind a real product, a fully generated image, a synthetic model wearing your garment. This is what disclosure rules are about.

Our position, and the one the book takes: shoot the real product consistently, use AI freely for work that removes or corrects things and be careful with work that invents things. Generated scenes routinely shift color, texture and scale in small ways, and those are the details shoppers use photos to judge.

Amazon: tag photorealistic AI-generated people

Per Amazon's Seller Central announcement dated July 22, 2026, sellers "must now tag product images that contain photorealistic AI-generated people with specific metadata." The Seller Central page blocks automated access, so we read it through a reputable summary that reproduces the notice (GeekSeller) and Quartz's July 24, 2026 report. Both describe it as in force from the announcement.

What the summary of the notice says:

  • Trigger: photorealistic AI-generated people (bodies, faces, hands) in product images, lifestyle images, product videos or A+ Content, in any of Amazon's worldwide stores.
  • How: add the keyword contains-synthetic-performer to the XMP dc:subject field with an IPTC-compatible metadata editor, before upload.
  • Not required: real people, even if altered with AI tools; characters from film, TV or games; images with no people; people who aren't photorealistic.
  • What shoppers see: Amazon says it may add an indicator that the image includes AI-generated people.

Quartz reports the change followed a New York law on synthetic performers in advertising that took effect in June 2026. For A+ Content, Amazon staff also point to "AI-generated" and "AI-generated people" checkboxes in the upload screen.

Separately, the main image still has to be a photograph of the actual product. AI doesn't change that.

Google Merchant Center: keep the IPTC tag

Google's Merchant Center help page on AI-generated content says: "All images created using generative AI must contain meta data indicating that the image was AI-generated." The field is IPTC DigitalSourceType, and the page names three values to preserve:

  • TrainedAlgorithmicMedia: created with a model derived from sampled content
  • CompositeSynthetic: a composite that includes synthetic elements
  • AlgorithmicMedia: created purely by an algorithm with no training data

It also says plainly: "Don't remove embedded metadata tags such as the IPTC DigitalSourceType property." This covers image_link, additional_image_link and lifestyle_image_link. Images from generative tools, including Google's own Product Studio, arrive tagged. Your job is to make sure no later step removes the tag. The page gives no effective date and doesn't mention C2PA.

Etsy: disclose in the listing description

Etsy's Creativity Standards (last updated June 10, 2025) place "seller-prompted AI creations" under the "Designed by a seller" category and say: "Sellers must disclose within their listing description if an item is created with the use of AI."

That rule is about the item. We found no Etsy rule written specifically about AI in listing photos, and Etsy's image requirements page doesn't mention AI. If you photograph a real handmade product and use AI only to clean the background, the written rule doesn't obviously apply. If your photos use generated scenes, a short line in the description costs nothing. Check the policy before you rely on either reading.

The EU AI Act: mostly not your obligation

Article 50 of the EU AI Act applies from August 2, 2026, with a grace period to December 2, 2026 for the marking duty on systems already on the market. The European Commission's FAQ puts the machine-readable marking duty on providers, the companies that build and sell generative AI tools. Deployers who publish deepfakes must label them visibly. Marking "does not apply when the AI system performs an assistive function for standard editing."

If you photograph real products, Article 50 doesn't add a new task to your shoot. The practical effect is upstream: more files from generative tools will arrive already marked, so a pipeline that keeps metadata matters even more.

Cross-channel policy table

Channel What triggers disclosure How to disclose Date checked
Amazon Photorealistic AI-generated people in images, video or A+ Content. Real people edited with AI don't count Keyword contains-synthetic-performer in XMP dc:subject; A+ checkboxes Oct 10, 2026 (via summary; Seller Central blocked)
Google Merchant Center Any image created with generative AI IPTC DigitalSourceType (e.g. TrainedAlgorithmicMedia, CompositeSynthetic); never strip it Oct 10, 2026
Etsy An item created with the use of AI A statement in the listing description Oct 10, 2026
EU (AI Act Art. 50) Falls on AI tool providers (marking) and deployers of deepfakes, not on standard editing Providers mark output; deepfakes labeled visibly Oct 10, 2026

[Image to add: The policy table as a Pinterest-friendly vertical graphic. Alt text: "AI product image disclosure checklist for Amazon, Google, Etsy and the EU, checked October 2026"]

How to tag and protect files in a batch

The metadata trap, from Chapter 13 of the book: image files carry IPTC and XMP text fields beside the pixels, and many resize tools (including the common "strip" options in libvips and ImageMagick) throw them away to save bytes. Strip the metadata and the disclosure is gone. The fix is a written policy plus a test.

ExifTool is the standard free tool for this. Run it after the last step that writes pixels.

1. See what your files carry now.

exiftool -r -XMP-dc:Subject -XMP-iptcExt:DigitalSourceType OUT/

2. Tag Amazon files that show AI-generated people. Put their file names in a plain text list, one per line, then:

exiftool -overwrite_original \
  "-XMP-dc:Subject-=contains-synthetic-performer" \
  "-XMP-dc:Subject+=contains-synthetic-performer" \
  -@ ai-people.txt

The remove-then-add pair stops a second run from adding the keyword twice. Leave out -overwrite_original the first time and ExifTool keeps a backup of each file.

3. Set Google's source type on generated or composite images. For a real product on a generated scene:

exiftool -overwrite_original \
  "-XMP-iptcExt:DigitalSourceType=http://cv.iptc.org/newscodes/digitalsourcetype/compositeSynthetic" \
  -@ ai-composite.txt

For a fully generated image, use trainedAlgorithmicMedia at the end of the address instead. Files from a generative tool may already carry the right value, so check before you overwrite.

4. Restore metadata a resize step removed. Copy it back from the master of the same name:

exiftool -overwrite_original -tagsFromFile MASTERS/%f.tif -XMP:all -IPTC:all OUT/google/

5. Find anything that slipped through.

exiftool -r -if 'not $XMP-iptcExt:DigitalSourceType' -p '$Directory/$FileName' OUT/google/

Then test the whole pipeline: run one known master through every step and check the final channel versions still carry the fields. Repeat whenever a tool or setting changes.

[Image to add: Terminal screenshot of step 1 output showing two files with the keyword and one without. Alt text: "ExifTool output listing XMP subject and digital source type for a folder of channel versions"]

The free sample includes the File naming worksheet, the first piece of the naming contract the scripts above rely on. Get the free sample.

FAQ

Do I need to disclose AI background removal? Not on any of these channels, as of October 2026. Amazon staff treat it like traditional editing, Google's rule covers images created with generative AI and the EU exempts standard editing assistance. Keep the image accurate to the product.

My model is a real person, retouched with AI. Does Amazon's keyword apply? No. The notice exempts real people, even if altered with AI tools.

Will Amazon reject an untagged image? The published summary doesn't describe penalties. It does say Amazon may show shoppers an indicator. Tag the files that need it and don't rely on enforcement details nobody has published.

What about C2PA Content Credentials? C2PA is an open standard that records where a file came from and how it was edited. None of the three channels requires it as of October 2026. Some editing software can write it; keep it if your files carry it.

Do these rules apply to eBay or Shopify? We found no published eBay rule on AI imagery. Shopify marks images from its own AI media tools with an invisible watermark. Check both before relying on generated images there.

Does the EU AI Act mean I can be fined for product photos? The Article 50 duties sit with AI tool providers and with deployers publishing deepfakes. Photographing real products isn't the target. If generated imagery is central to what you sell in the EU, take proper advice.

The takeaway

Shoot the real product, keep AI to correcting what you shot and, when AI does create something, say so in the place each channel reads. The book's Chapter 13 and its AI chapter cover the full metadata policy. Start with the free sample.

The Book + Template Pack ($29.99) adds all 14 chapters, the channel spec sheet and the export scripts.

This is general information, not legal advice. Check each channel's current policy.

Sources

Amazon's Seller Central announcement blocks automated access; the Amazon rule is taken from the GeekSeller summary that reproduces it, Quartz and an Amazon staff reply on Seller Forums.

Jeremy Maz

Image maker/artist/photographer/creative etc. based in Los Angeles, CA. Bachelor's of Fine Arts in Photography + Imaging from Art Center College of Design in Pasadena, CA (I paid a lot of money to be able to say that so don't judge).

https://www.jeremymaz.com
Previous
Previous

Google Merchant Center image requirements: the 500 × 500 px rule

Next
Next

Product image margins: how much frame fill is right?