Amazon's AI Image Disclosure Tag: What Sellers Must Do Now
Amazon now requires a metadata keyword on product images that contain fully AI-generated people. Here is exactly what triggers it, the three ways to add it, and why you must verify the tag right before every upload.
On 22 July 2026, Amazon told sellers to identify product media that contains photorealistic people generated entirely by AI. The rule sits in Amazon's official product image requirements (Seller Central login required). The mechanism is a metadata keyword — contains-synthetic-performer — written into the image before upload.
If you use AI-generated models in your listing images, this affects you now. We built a free bulk tagging tool that writes the keyword and verifies it, entirely in your browser, with no signup and no uploads.
About this article. It is compiled from Amazon's product image requirements in Seller Central, from the text of New York General Business Law § 396-b (S.8420-A / A.8887-B), and from our own testing of how the tag behaves in real workflows. Legislation references and dates are cited so you can check them. It is not legal advice — Amazon's own page is the authority, and policy can change.
The rule in one paragraph
Before you upload an image that contains a photorealistic human who was created entirely by AI, use an IPTC-compatible metadata editor to add the keyword contains-synthetic-performer to the image's XMP metadata, in the dc:subject field. In Amazon's words: "If you include this metadata, Amazon will add a disclosure to inform audiences that the image includes AI-generated people, where applicable."
The official page covers product images; press coverage of the announcement reported that A+ content is in scope too, so treat AI people in A+ banners the same way. Amazon also publishes a dedicated help article, How to tag media that contains an AI-generated person, linked from the image guide.
What triggers it — and what does not
This is where most sellers get it wrong in both directions. The test is not "did I use AI" and it is not "is there a person."
| Your image | Tag it? |
|---|---|
| An AI-generated model wearing or using your product | Yes |
| An AI-generated bystander in the background of a lifestyle scene | Yes |
| A+ content or a banner featuring an AI-generated person | Yes |
| A real photo of a real person, even heavily retouched with AI | No |
| A real model whose background was replaced by AI | No |
| Product-only images with no people at all | No |
| Cartoons, illustrations, stylised mascots | No |
| Characters from films, games or similar works | No |
The line is photorealistic + entirely synthetic + not a recognisable real person. New York's statute defines a synthetic performer as a digital asset created, reproduced or modified by computer using generative AI or a software algorithm, intended to give the impression of a human performance by someone who is not an identifiable person.
Why the law exists
Amazon did not invent this on its own. New York's synthetic performer disclosure law (General Business Law § 396-b, from bills S.8420-A / A.8887-B) took effect on 9 June 2026. It requires disclosure when a synthetic performer stands in for a real person in commercial advertising. Penalties run $1,000 for a first violation and $5,000 for subsequent ones, with a short window to remedy. Amazon's metadata requirement is how a marketplace operationalises that law across millions of listings.
Three ways to add the tag
1. A bulk tool (fastest)
Drag your images into our free tagging tool. It writes the keyword into XMP dc:subject, mirrors it into the Windows "Tags" field so you can spot-check it in File Explorer, and then re-reads each output file to confirm the keyword actually landed. Everything runs locally in your browser — the images never leave your device. It also converts WebP to JPG, because Amazon does not accept WebP.
2. ExifTool (most precise)
exiftool -xmp:dc:subject+=contains-synthetic-performer image.jpg
The += appends rather than replacing, so any existing keywords survive. Run it on a folder to process a batch.
3. The operating system (least reliable)
On Windows, right-click the file → Properties → Details → type the keyword into Tags. On macOS, open the image in Preview → Tools → Show Inspector → Keywords.
Both work, but neither writes standard XMP consistently across formats, so always verify afterwards rather than assuming.
The part almost everyone misses: tag last, verify last
Metadata is fragile. Between the moment you tag a file and the moment it reaches Amazon, it usually passes through several steps — retouching, compression, resizing, a format conversion, an asset library, an agency portal. Many of those strip metadata on export.
A useful rule of thumb from our testing:
- Strips the tag: re-encoding of any kind — compression tools, "save for web" exports, resizing, format conversion, most social and chat apps
- Keeps the tag: copying the file, zipping it, sending it as a file attachment, uploading it as-is
So the order matters:
- Finish the image completely — all retouching, resizing, cropping, format conversion
- Then write the tag
- Verify immediately before uploading
- Keep the verified file as the master you upload from
If you only remember one thing from this article: tagging is not a step you do once at the start. It is the last thing you do before upload.
How to verify
The strict check:
exiftool -xmp:dc:subject image.jpg
The output should contain contains-synthetic-performer. On Windows you can also right-click → Properties → Details and look at the Tags row, though whether it appears there depends on your Windows version and installed codecs. XMP is what Amazon reads, so trust ExifTool over File Explorer.
Our free tool does this check for you automatically: after writing each file it re-reads the XMP from the output and only reports success once the keyword is confirmed on disk.
Two related traps while you are in there
WebP is not accepted. Amazon's supported formats are JPEG (recommended), TIFF, PNG and non-animated GIF. WebP is not on the list, so if your AI tool exports WebP, convert to JPG — a tagged WebP is still a rejected upload.
Check your dimensions at the same time. Amazon's spec is 500 to 10,000 pixels on the longest side, and JPEG is the recommended format. Since you are touching every image anyway, it is a cheap moment to catch the ones that fall outside that range. Upscaling only adds blur — re-export from the source instead.
If you generate images with HedaAI
Images produced with an AI model from the built-in cast get the disclosure tag written automatically when you download them — at download time specifically, after watermarking and upscaling, so nothing in the pipeline can strip it afterwards. Product-only images with no people are never tagged, and you can turn the behaviour off in Account settings if you tag through your own workflow.
A five-step checklist
- Work out which of your live listings use AI-generated people
- Fix the source: make tagging the final step of your image workflow, not the first
- Tag the images you are about to upload, in bulk
- Verify each one immediately before upload — do not assume the tag survived
- Re-upload for existing listings you want covered; local edits alone change nothing on Amazon