In July 2026 Amazon began asking third-party sellers to identify product images and videos containing photorealistic people generated entirely by AI. Qualifying files must carry a machine-readable keyword in their metadata before upload, and Amazon indicated it will surface a disclosure to shoppers where the rule applies.
This is a labeling requirement, not a ban. Most product sellers are unaffected. It is still worth ten minutes of your time, because the penalty for getting it wrong lands on the listing rather than the image.
Who this applies to
The rule covers photorealistic people created entirely by AI, appearing in listing images, video or A+ Content. It does not cover:
- A generated image of your product with no person in it
- A 3D render of your product
- A real photographed model with a release
- Licensed stock photography of a real person
The grey zone is composites: real photography with AI-generated faces, bodies or poses substituted in. Classify those carefully rather than assuming they pass.
How the tagging works
Amazon asks for an IPTC or XMP metadata keyword inside the file itself, rather than a checkbox in Seller Central. In practice that means writing the keyword into the image file before you upload it, using whatever your team already uses to handle metadata: Photoshop's File Info panel, Bridge, ExifTool, or your DAM.
Two practical notes. First, metadata survives some export pipelines and not others, so verify the tag after export rather than before. Second, agree internally on who applies it, because the person generating the image and the person uploading the listing are often different people.
Build a simple audit
Sort every listing asset into four buckets:
- Real photography with a shoot record and a model release.
- Licensed stock of a real person, with a licence that covers marketplace use.
- Composite imagery where generated elements were combined with real photography.
- Fully synthetic person created by a model or tool.
Bucket four needs the tag. Bucket three needs a judgement call and a written record of why you decided as you did. Buckets one and two need nothing but the paperwork you should already hold.
Keep source files for everything. If Amazon ever questions an image, the fastest resolution is showing where it came from.
The part nobody puts in the policy summary
Amazon said it will show a shopper-facing indicator on media covered by the rule. Research on labeled AI content generally points the same way: disclosed AI creative tends to draw lower engagement than undisclosed equivalents.
So the practical question is no longer only whether you may use an AI-generated person. It is whether you want an AI label sitting on your best lifestyle image while a competitor's identical scene carries none.
The straightforward way around it
Two options avoid the rule entirely.
Shoot real people. Slower and more expensive, still the gold standard for emotive brand imagery, and no disclosure question arises.
Build the scene around the product. Most lifestyle listing images do not need a face. A kitchen counter, a desk, a gym floor, hands out of frame. Built from your product's 3D model, these scenes are accurate, reusable across colorways, and carry nothing to disclose.
We take the second route for most client work, which is why our lifestyle scenes can be re-rendered for a new color or a new season without going back to a studio or a model release.
If you are auditing your listings and want the imagery rebuilt properly, start with a $130 test render and see how the product looks before committing anything.
See this quality on your own product
First-time clients get a complete, high-resolution 3D render of their product for $130. Judge the quality before committing to a full package.
