COMMERCE · Platform Updates

Amazon now wants a hidden metadata tag on every photorealistic AI face in your product images

A New York disclosure law took effect in June. Amazon just turned it into a platform rule, with a specific metadata keyword, worldwide.

Chandni Melwani

Chandni Melwani

Founder & Editor

Jul 24, 2026 · 1 MIN READ

The News

On July 23, 2026, Amazon told third-party sellers to label product images and videos that contain photorealistic AI-generated people. Sellers must add the keyword 'contains-synthetic-performer' to the dc:subject field of an image's XMP metadata, using an IPTC-compatible editor, before uploading. The rule covers main listing media and A+ Content across Amazon's stores worldwide, and follows New York's synthetic-performer disclosure law, which took effect June 9, 2026.

Know More

  • XMP is the metadata layer inside an image file — the same place a camera writes shutter speed and a photographer writes copyright. It travels with the file and is invisible on the page.
  • IPTC is the publishing standard that defines those fields. An IPTC-compatible editor is anything that can write them: Photoshop, Lightroom, ExifTool and most asset-management systems.
  • Where it applies: main listing media and A+ Content, across Amazon's stores worldwide.
  • Who bears the penalty: the business that creates the ad — $1,000 for a first violation, $5,000 for each after. Platforms that merely distribute are exempt, which is precisely why Amazon has pushed this onto sellers.
  • Excluded, in Amazon's own list: real people, even where AI tools altered them; characters from films, TV, streaming, documentaries, games and other expressive works; images with no people at all; and people who are not photorealistic. A stylised or illustrated AI figure falls outside the rule, as does an AI-edited background with no synthetic human.
  • What shoppers see: Amazon says that where the metadata is present it will add an indicator telling customers the image includes AI-generated people. The tag is the seller's input; the visible disclosure is rendered by Amazon.
  • Enforcement: the notice attaches no penalty of its own. Amazon's general image policies already allow it to remove non-compliant images or suppress listings, and that is the lever available — not a stated consequence of a missing tag.
  • Not published: any grace period, any audit or detection method, and whether Amazon will check the tag or rely on sellers.

Amazon’s newest seller rule lives somewhere no shopper will ever look. Every image file carries a hidden metadata layer — the place a camera records shutter speed and a photographer records copyright — and since July 23, any listing photo or video showing a photorealistic AI-generated person has to carry the keyword contains-synthetic-performer written into that layer before upload.

The rule is Amazon’s way of absorbing New York’s synthetic-performer disclosure law, which took effect June 9 and puts the duty to disclose on the business that creates an ad, not the platform that runs it. Rather than police that state by state, Amazon pushed the requirement down to sellers and applied it across its stores worldwide.

For sellers who went early on AI-generated models in their photography, tagging is the easy half. Finding every affected file is not. A brand with a few hundred products can have synthetic faces scattered through hero shots, comparison modules and the long-form brand pages Amazon calls A+ Content, much of it made months ago by agencies who have since moved on. One missed image is enough to put a listing at risk of suppression.

The gap between an AI tool generating a face and that image arriving in Seller Central correctly tagged is, for now, bridged by human memory. That is a builder’s opening for provenance tooling, and a preview of a rule likely heading to every platform where AI creative touches a buy button.

#Amazon#Synthetic Media#AI-Generated Content#Marketplace Compliance#Product Imagery#Platform Updates#E-commerce

Frequently Asked Questions

Why put the disclosure in metadata rather than on the image?

Because metadata travels with the file. It survives copying, re-uploading and passing between systems, whereas a watermark can be cropped and a caption rewritten, and it stays machine-readable for the kind of audit a regulator or a platform would actually run. It is not a substitute for telling the shopper, though — Amazon says that where the tag is present it adds its own indicator on the listing. Sellers supply the metadata; Amazon decides what the customer sees. The trade-off is that a seller cannot tell from the page whether a given file was ever tagged.

Does this apply outside New York?

Yes. New York's law is the trigger, but Amazon applied the requirement across its stores worldwide rather than trying to enforce it state by state. If your listing shows an AI-generated person, the tag is expected regardless of where you sell.

Who is liable, the seller or Amazon?

The business that creates the advertisement. New York's law puts the disclosure duty on the advertiser, not on platforms that merely distribute the ad, which is exactly why Amazon is pushing the requirement down to sellers. Penalties run $1,000 for a first violation and $5,000 for each one after.

What counts as a "synthetic performer"?

A photorealistic human figure generated entirely by AI. New York's statute defines a synthetic performer as a digitally created asset made or modified by generative AI that is meant to read as a human performance by someone not recognisable as any identifiable real person. Amazon's version excludes real people whose appearance was AI-altered, characters from films, television, streaming, documentaries or games, images with no people, and figures that are not photorealistic. Background removal without a synthetic human is not the target.

What happens if a seller does not tag?

Amazon's notice does not say. It sets out no grace period, no detection method and no penalty specific to the tag. What exists is the general image policy, which already lets Amazon pull non-compliant media or suppress a listing — so the exposure is real but undefined. For large catalogs the harder problem is upstream anyway — finding every affected asset, not tagging it.

Chandni Melwani

Chandni Melwani

Chandni Melwani is the founder and editor of New in AI, covering AI agents, M&A, and enterprise adoption. She holds a Master's in Management of Artificial Intelligence from Queen's University and brings a practitioner's perspective from her work in Data and AI leadership.

Follow ↗

Commerce Briefing

Get the two-beat briefing before the market opens.

One email. Corridor capital moves and Commerce platform shifts, decoded daily.

No spam. Unsubscribe anytime.

More from Commerce