What changed: Amazon now requires you to tag any product image, video or A+ module that contains a photorealistic person generated entirely by AI. The tag is a metadata keyword you embed in the file before you upload it, and Amazon adds a shopper-facing indicator to listings that carry it. Amazon sent the notice to sellers on Wednesday 22 July 2026 and the requirement is already in force — there is no future deadline to prepare for. It is driven by a New York disclosure law, but it applies to how you upload to Amazon regardless of where you sell from.

What changed

Amazon notified third-party sellers on 22 July that images and videos containing photorealistic AI-generated people must be labelled with machine-readable metadata before upload. CNBC reported the notice on 23 July and Forbes covered it independently on 25 July.

The mechanics:

  • The keyword is contains-synthetic-performer, added to the dc:subject field of the file’s XMP metadata using any IPTC-compatible metadata editor. It has to be there before the asset goes into your listing or A+ content.
  • It covers images, video and A+ content — not just the main image.
  • Amazon adds a disclosure indicator on the detail page telling customers the image includes AI-generated people.
  • It applies only to fully synthetic photorealistic humans. Real people whose photos were AI-edited or retouched are not covered. Neither are characters from film, television or video games, non-photorealistic illustrations, or images with no people in them.

The driver is New York’s synthetic-performer disclosure law — an amendment to General Business Law § 396-b that took effect 9 June 2026, requiring advertisers to disclose when a digitally created figure indistinguishable from a real person stands in for a human actor. Penalties under that statute are $1,000 for a first violation and $5,000 for each subsequent one.

Note what that means: the dollar penalties are a New York State penalty on the advertiser, not an Amazon fee. Amazon’s own lever is listing and account enforcement — and Amazon has not published what that looks like here.

What it means for a private-label seller

The AI model shot just became a compliance decision instead of a creative one.

If you sell in beauty, skincare, apparel or supplements, you are the most exposed, because those are exactly the categories where a generated model replaced a $2,000 photoshoot over the last two years. A lot of sellers do not actually know whether their lifestyle images contain a real person. If you bought imagery from a freelancer or a creative agency, the honest answer is that you have to go and ask.

The trade-off nobody will tell you about: tagging correctly is cheap and it is the right call — but it puts an AI-disclosure indicator on your detail page, sitting next to competitors who used a real model and carry no indicator. Nobody has measured what that does to conversion yet, and in skincare, where the image is the credibility, it may not be nothing. Not tagging is the option that looks cheaper this quarter and carries both the state-law exposure and whatever Amazon decides enforcement means. Tag it, and treat the indicator as a reason to reconsider the asset — not as a reason to leave it untagged.

The bright line in the policy is blurrier in practice. “Generated entirely by AI” and “a real person, AI-edited” are clean categories in a help doc. A composite — a real model’s face on a generated body, or a real shoot with a synthetic second person in frame — is the common case in beauty imagery, and the policy as reported does not clearly address it.

What to do about it

  1. Audit which live ASINs use fully AI-generated people — main image, secondary images, A+ modules and video. This is a catalogue-wide check, not a hero-image check.
  2. Ask your creative supplier in writing whether any delivered asset contains a synthetic person. If they used a generative tool for models, you need that answer on record.
  3. Tag the files that qualify. Add contains-synthetic-performer to the XMP dc:subject field, then re-upload. macOS: Preview → ⌘I → Keywords. Windows: file Properties → Details → Tags. Command line: exiftool -XMP-dc:Subject+=contains-synthetic-performer yourfile.jpg.
  4. Do not tag images of real people to be safe. Amazon explicitly scopes this to fully synthetic humans. A disclosure indicator on a real photoshoot misrepresents your own listing.
  5. If you would rather not carry the indicator, change the asset — a real shoot, or a clearly non-photorealistic illustration. The requirement is scoped to photorealistic synthetic humans, so an obviously illustrated figure falls outside it.

What this doesn’t tell you

Four things are genuinely unknown right now, and anyone telling you otherwise is guessing:

  • What the indicator does to conversion. No data exists yet. It is days old.
  • How Amazon enforces it. Amazon has not published a consequence for failing to tag.
  • What happens to imagery already live. The instruction is to tag before uploading. Amazon has not said what it expects for assets uploaded before 22 July.
  • Exactly when the shopper-facing indicator appears. Forbes noted the criteria have not been detailed publicly.

The full policy detail sits in Seller Central help page GFXHCHYZRGJRBZA5 (“how to tag media that contains an AI-generated person”), which is behind the Seller Central login — worth reading in your own account before you change anything.

Sources

The core facts here — the requirement, the metadata keyword and field, the exclusions, the New York statute and its penalties — are confirmed by two or more independent outlets. Where something is not yet known, this post says so rather than filling the gap.