If you've uploaded a product photo with an AI model lately and watched it get kicked back — or quietly vanish from your listing — you're not imagining things. Amazon has a labeling rule now that catches a lot of sellers off guard, mostly because nobody explains it in plain English.
Short version: if your image or video contains a photorealistic AI-generated person, you have to tag the file with a specific metadata keyword before you upload it. Skip the tag and Amazon can reject the upload, strip the image, or flag the listing.
This isn't a ban on AI images. It's a disclosure rule. And once you know the one keyword that matters, it's a five-minute fix.
Why Amazon cares about AI people specifically
The trigger is a law. New York's Synthetic Performers Act (signed December 2025, in force since June 2026) says any ad using a "synthetic performer" — a digital human that isn't a real, identifiable person — has to say so conspicuously. First violation runs $1,000, each repeat $5,000.
Amazon sells into New York, so it built the requirement into Seller Central for every marketplace, not just the US. The company is also a member of the C2PA coalition, which is why metadata — not just a checkbox — is what its systems actually read.
Two official spots sellers should bookmark:
- Amazon's Product Image Guide, which now carries a section on disclosing AI-generated people.
- The Seller Forums announcement that lays out the
contains-synthetic-performerrequirement.
The one keyword you need to know
The tag is literally:
You add it to the image's XMP dc:subject field (that's the keywords/subject metadata). When Amazon sees it, it shows a disclosure to shoppers telling them the person isn't real.
A few things the official guidance is clear about:
- You only need this for photorealistic AI people — fully generated humans, not based on a real person.
- You do not need it if the image shows only real people (even if you touched them up with AI), only fictional characters from movies or games, no people at all, or people that aren't photo-realistic.
- Plain product shots, AI background swaps, and AI scene generation with no human figure? Those go through the normal workflow. No file-level tag required.
That last point matters, because a lot of the panic online conflates "AI image" with "must be labeled." It doesn't. Only the synthetic person does.
How Amazon actually catches you
Two layers, roughly:
1. Metadata check. Amazon reads the file header — software name, timestamps, copyright, and any C2PA provenance. If the tool that made the image wrote its name (Midjourney, DALL·E, Stable Diffusion) and there's no disclosure tag where one should be, the system flags it.
2. Pixel-level scan. Deep-learning models look for generative patterns in texture, lighting, and perspective — the statistical "tells" AI images leave even after you retouch them by hand.
In seller tests, AI scene images uploaded with no metadata at all had close to a 0% pass rate. Some slipped through at first and got pulled later in a sweep. So "I'll fix it after it goes live" is a losing bet.
How to label your images (step by step)
1. Keep the original generation metadata.
Don't strip it on export. Tools like Adobe Firefly, DALL·E, and TikTok Symphony embed provenance automatically. When you upload AI people through Amazon's own A+ creative asset tool, it can apply the tag for you if you pick "AI-generated person." Leaving the data alone is the path of least resistance.
2. Add the tag manually when metadata is gone.
For images that lost their tag somewhere in editing:
- Adobe Photoshop: use the File Info / metadata panel to add
contains-synthetic-performerto the subject or keywords field. - ExifTool: the command-line standard. One line writes the XMP field directly.
A clear contains-synthetic-performer string is what the detectors look for. Don't get cute with synonyms.
3. Verify before you upload.
Open the final file in a metadata viewer — online or local — and confirm the tag survived. Compression and format conversion silently drop metadata, so check the exported file, not the source.
4. Declare older assets when Amazon asks.
Amazon is adding a declaration entry in Seller Central. Once it's live for your marketplace, proactively declare historical AI people so you're not dinged for uploads from before the rule.
A quick compliance sweep for your catalog
Worth doing this week, not next quarter:
1. Flag every image or video with an AI-generated person.
2. Confirm each one carries the contains-synthetic-performer tag.
3. Don't skip A+ content and product videos — that's exactly where this rule bites hardest.
4. Build a one-line SOP: tag AI people before upload becomes a mandatory step, like naming the file.
A suppressed listing costs more than the two minutes it takes to tag an image. I've watched best-sellers flatline for a week over a missing keyword.
FAQ
Do I need to label every AI image on Amazon?
No. Only images and videos that contain a photorealistic AI-generated person. Real people (even AI-edited), non-human characters, and images with no people don't need the tag.
What exact metadata does Amazon want?
The keyword contains-synthetic-performer in the XMP dc:subject field. Amazon also reads C2PA provenance where present.
Can Amazon detect AI people I edited by hand?
Yes. Beyond metadata, pixel-level analysis catches generative patterns in lighting, texture, and perspective — retouching doesn't erase them.
What happens if I don't label them?
The image can be rejected at upload, removed after publishing, or trigger a listing or account flag. Worst case the listing goes offline.
Does this apply to A+ content and videos?
Yes — and that's where enforcement is tightest right now. Plain main product photos without people aren't subject to the file-level tag.
Key takeaways
- Amazon now requires a
contains-synthetic-performermetadata tag on images and videos with photorealistic AI-generated people. - It's a disclosure rule tied to New York's Synthetic Performers Act, enforced globally — not a ban on AI imagery.
- Keep original tool metadata; repair missing tags with Photoshop or ExifTool; verify before upload.
- Audit A+ content and videos first. Make tagging a standard step.
