Amazon's AI image disclosure rule
On 22 July 2026 Amazon told third-party sellers that listing images, A+ content and product video containing photorealistic people generated entirely by AI must carry a specific metadata tag before upload. The requirement applies across all worldwide stores, and it stems directly from New York's synthetic performer disclosure law.
What is required
- Use an IPTC/XMP-capable editor to write the keyword
contains-synthetic-performerinto the image's XMPdc:subjectfield - It has to be done before upload
- Amazon reads the keyword and surfaces a disclosure to shoppers on the detail page
For the field details and three free ways to tag a single file, see the contains-synthetic-performer page.
Scope and exemptions
| In scope | Exempt |
|---|---|
| Photorealistic people generated entirely by AI | Real people, even when the image was edited with AI |
| AI-generated photorealistic bystanders in the background | Fictional characters from film, television or games |
| Images and video inside A+ content | Cartoons, illustrations and other non-photorealistic figures |
| Product video | Images with no people in frame |
What about listings that are already live
In its seller forum reply, Amazon stated that compliance is mandatory going forward for new uploads; images already online fall under a best-practice recommendation to update them while editing, with no retroactive deadline set.
One practical detail is easy to miss: changing the local file does nothing to a live listing. Getting a tag onto an existing image means re-uploading it. So the pragmatic approach is to sort listings by traffic and fold tagging into the work you were already going to do — routine image edits, seasonal refreshes — rather than re-uploading an entire catalogue for its own sake.
The law behind it
New York General Business Law § 396-b (S.8420-A / A.8887-B), effective 9 June 2026, requires commercial advertisements to disclose when a “synthetic performer” stands in for a human one. The law defines a synthetic performer as a digital asset created, reproduced or modified by a computer using generative AI or a software algorithm, intended to give the impression of a human performance by someone who is not recognisable as any identifiable real performer.
- Civil penalty of $1,000 for a first violation and $5,000 for each one after
- Publishers and distributors notified of an undisclosed synthetic performer have five days to remove the content or bring it into compliance
Note that the statute binds advertisers generally, not just Amazon — the same obligation follows ads running in New York on other channels. Amazon is simply the first channel to turn it into an enforced platform workflow.
Questions sellers actually ask
Can a main image use an AI model at all?
Yes — the rule is about disclosure, not prohibition. Amazon has not banned AI-generated people in listing images; it requires that they be declared. Main images still have to meet the existing standards separately: pure white background, product filling at least 85% of the frame, 1,000 pixels or more on the longest side.
What happens if a listing is already live with an untagged AI image?
Amazon's stated position is that compliance is mandatory for new uploads, with older images a best-practice recommendation and no retroactive deadline. Editing your local copy changes nothing on a live listing — the tag only reaches Amazon through an upload. The practical approach is to fold re-tagging into work already scheduled: routine image refreshes, seasonal updates, A+ rebuilds.
How would I know if this is hurting a listing?
You would not, directly. Suppression of a main image shows up as lost impressions rather than a notification, which is what makes it worth checking proactively rather than waiting for a signal. If traffic to a listing drops without an obvious cause and its images contain AI-generated people, read the tag on the file you uploaded before looking further afield.
Where this sits among the other AI disclosure rules
Amazon is one channel implementing one law. The same creative running elsewhere can face different requirements at the same time:
| Channel | Mechanism |
|---|---|
| Amazon | Metadata — you write the tag, Amazon renders the disclosure |
| Google Ads / YouTube | A setting in the ads product, plus visible overlays for EU, India and New York campaigns |
| Meta | Meta's own AI info label, applied when it detects or is told |
| TikTok | A toggle per ad, reset when a campaign is duplicated |
Only Amazon reads a file-level tag. Everywhere else the disclosure lives in the ad platform or on the creative itself — see the label generator and the EU rules from 2 August 2026.
Seller action list
- Write down your judgement criteria first. Deciding which images need the tag is the prerequisite for everything else
- Move tagging upstream into production. Tag at the point of generation rather than retrofitting a whole library later
- Verify before upload. Compression, export, asset managers and agency portals all strip metadata — drop the file into the checker to confirm the tag survived
- Keep creative records. The New York statute turns on knowledge, so briefs and prompt logs cut both ways: they evidence diligence and they evidence awareness
- Reconsider the creative route. AI-labelled content costs some consumer trust; using AI for backgrounds and scenes while keeping human figures real avoids the requirement at source
This page summarises publicly available information and is not legal advice. Amazon's official announcement and the New York statute govern.