Amazon Now Requires a Metadata Tag on AI-Generated People in Listing Images

Amazon's instruction is specific. Sellers must use an IPTC-compatible metadata editor to add the keyword contains-synthetic-performer to the dc:subject (XMP) field of the file before uploading it. Amazon then displays a disclosure to shoppers indicating that the image includes AI-generated people, where applicable.

In its notice to sellers, Amazon explained that recent legislation requires disclosure when images or videos in advertisements contain photorealistic AI-generated people. The company has published a Seller Central help page, "How to tag media that contains an AI-generated person," covering the technical steps.

The requirement covers four content types:

• Main and secondary product images
• Lifestyle images
• Product videos
• A+ Content modules

The scope is narrow but the definition is precise. It applies to people who were entirely generated by AI and appear photorealistic — a synthetic model who does not correspond to any real person.

This is not an Amazon quality initiative. It is a compliance response to a state law.

New York's synthetic performer disclosure law took effect on June 9, 2026. It amends General Business Law § 396-b (S.8420-A / A.8887-B, signed in December 2025) and requires disclosure whenever a synthetic performer — an AI-created digital asset designed to give the impression of a human performance — appears in an advertisement in any medium.

The penalties are modest per incident but scale quickly:

• $1,000 for a first violation.
• $5,000 for each subsequent violation.
• A five-day window for publishers and distributors to come into compliance once they receive notice.

Amazon distributes advertising at enormous scale, so pushing the tagging obligation down to the sellers who create the creative is the only workable path. That is also why the rule is worldwide rather than New York-only: Amazon is not going to maintain a separate creative pipeline per state.

Amazon is unusually clear that over-tagging is also wrong. The keyword should not be added in four situations:

• Images containing only real people — even if those photos were edited or enhanced with AI-powered tools.
• Characters from movies, television, streaming content, documentaries, video games, or other expressive works.
• Images that contain no people at all.
• People who are not photorealistic — illustrations, cartoons, stylized renders.

Run your catalog through that list and the practical answer for most sellers is: no tag required. A packshot on a white background has no people in it. A hand model holding your product is a real person. An upscaled or background-corrected photograph of your actual product is still a photograph of your actual product.

The sellers who need to pay attention are the ones generating synthetic models — apparel and accessories brands using AI-generated people to show fit, home goods sellers generating lifestyle scenes with people in them, and anyone whose A+ Content includes AI-created lifestyle photography with human subjects.

The exemption for real people, even where they have been altered with AI-powered tools, is broader than most sellers expect. If you photographed a real model and then used AI to clean up the background, correct the lighting, or upscale the resolution, the person in the frame is still that real person. No tag.

The line is whether the person exists. Generate a face from a prompt and you have created a synthetic performer. Retouch a face that a camera captured and you have not. Composite work — a real model's head on a generated body, for example — sits closer to the synthetic side. If you are genuinely uncertain and your case is not one of the four explicit exclusions above, tagging is the lower-risk choice.

The tag lives inside the file, which means every step of your image pipeline can destroy it. XMP metadata is routinely stripped during exports, resizing, format conversion, and by asset-management systems. A perfectly compliant master file can easily become a non-compliant upload.

The fix is ordering. Tag last, then verify.

Sellers who fail this rule will overwhelmingly fail it by accident. They will tag the master file correctly in a desktop editor, then run a batch resize or a WebP conversion on the way to upload, and ship an untagged file with a completely clean conscience.

If you cannot restructure your pipeline so that tagging comes last, at minimum audit the output. Pull three files at random from what you actually uploaded last week and check whether the metadata you set is still in them. The answer surprises most people the first time.

There is a commercial dimension here that Amazon's help page will not mention. Disclosure labels change buyer behavior.

Research by Khalil Israfilzade published in 2025 found that AI-labeled advertisements generated lower consumer trust and lower purchase intent than advertising labeled as human-made, with the gap widest for high-involvement products — the considered, higher-price purchases where shoppers scrutinize a listing hardest.

That does not make the tag optional. It does make the underlying creative decision worth revisiting. If you sell apparel, furniture, or anything else in a high-consideration category, a real photograph of a real model may now outperform a synthetic one for reasons that have nothing to do with image quality — and it carries no tagging obligation at all.

Using AI on the environment rather than on people keeps you outside the rule entirely. Background replacement, upscaling, and lighting correction applied to genuine product photography require no disclosure, because nothing in the frame was invented.

1. Filter your catalog for any listing image, video, or A+ module that shows a person. For most catalogs this is a short list.
2. For each one, answer a single question: was this person captured by a camera, or generated from a prompt? Camera means no tag.
3. For the generated ones, restructure the workflow so that metadata tagging is the final step, after all resizing, conversion, and compression.
4. Add contains-synthetic-performer to dc:subject with an IPTC-compatible editor, then reopen the exported file and confirm the keyword is still there.
5. Re-validate the finished files against Amazon's main-image specs — background purity, framing, and resolution are still the things that get listings suppressed.
6. Read Amazon's Seller Central page "How to tag media that contains an AI-generated person" for the current technical steps, since the exact trigger conditions for the shopper-facing indicator have not been fully published.