Amazon AI-Generated People in Images: 2026 Disclosure Rules

What did Amazon change, and when did it start?

In late July 2026 Amazon told sellers that any listing image, video or A+ content showing a photorealistic person created by AI must carry a metadata keyword, contains-synthetic-performer, before upload. Amazon uses that tag to show shoppers an AI indicator where it decides one applies. The rule covers all of Amazon's stores worldwide.

The trigger was a New York statute. Governor Kathy Hochul signed the synthetic performer bill on December 11, 2025, and it took effect on June 9, 2026, as an amendment to New York General Business Law section 396-b. About six weeks later, Amazon posted a Seller Central news item. CNBC reported on July 23, 2026 that sellers had been told the day before, and a copy of the notice reproduced by several seller tool providers is dated July 22, 2026.

Three details in the notice matter most for day-to-day work:

• Where the tag goes. Inside the file, in the XMP dc:subject field, written with an IPTC-compatible metadata editor. Not in the title, bullets, file name or alt text.
• When. Before the file is added to a listing or an A+ module. Amazon reads the tag at upload.
• What shoppers see. Amazon says it will add an indicator "where applicable." It has not published the criteria for when that label appears, a gap CNBC also pointed out.

One caveat on sourcing. The full Seller Central announcement and the linked help page, titled "How to tag media that contains an AI-generated person," sit behind a seller login. What we describe here comes from CNBC's report and from the notice text that seller software companies republished. If your account shows different wording, follow your account.

What does New York's synthetic performer law actually require?

Section 396-b says a business that produces an ad for its goods or services must clearly disclose any synthetic performer it knows is in that ad. A synthetic performer is a digital human, built with generative AI or other software, that is not any identifiable real person. Fines run $1,000, then $5,000.

Reading the statute on the New York State Senate site alongside summaries from Cooley, Skadden and Hunton, these are the parts a brand owner should know:

• It is about fake humans, not AI in general. Retouching a real model, generating a background or writing copy with AI does not trigger the duty. A fully generated person who looks real does.
• Any software counts. Cooley notes the definition covers assets built with a "software algorithm," so a digital human made with traditional 3D or VFX tools can also qualify, not just one from an image generator.
• Knowledge matters. The duty applies when the advertiser has actual knowledge. That makes your creative records important, because "we did not know" is hard to argue if your own brief asked for an AI model.
• "Conspicuous" is not defined. The law does not say what wording, size or placement the disclosure needs. On Amazon, the platform's indicator is the mechanism sellers have been given.
• Exemptions. Ads for films, TV, streaming shows, documentaries and video games are exempt when the synthetic character is used the same way as in the work itself. Audio-only ads are exempt, as is AI used only to translate a real performer's language. Publishers and platforms that simply carry the ad are not liable under this section.

Two more points of context. Skadden and Cooley both flag that a December 11, 2025 federal executive order on state AI rules could lead to challenges, so the law's future is not settled. And the statute applies to ads in New York, while Amazon chose to apply its tagging rule in every store. This article is general information, not legal advice; confirm how the law applies to your business with your own counsel.

Which images and videos need the tag, and which do not?

Tag an asset when it shows a person who looks photographic and was generated, rather than photographed. Skip the tag when every person shown is real (even if AI edited them), when the only characters come from films, TV or games, when no person appears, or when the people are clearly illustrated or stylized.

Because Amazon applies this in all stores, a UK brand selling on amazon.co.uk or an Indian exporter selling on amazon.com is in scope even though neither is based in New York. The table below covers the cases we see most often in listing and A+ work.

The gray zones are partial figures, heavily composited images where a real face sits on a generated body, and 3D renders of people. Amazon's public notice does not cover these, so we tag them and keep a note of why. Over-tagging means a shopper may see a label you could have avoided. Under-tagging is the bigger risk if the law or Amazon later looks at the asset.

How do I audit an existing catalog for AI-generated people?

List every image, video and A+ module on your live listings and record where each came from: a photo shoot, an AI tool or both. Anything showing a generated person joins a re-tag queue. Tag those masters, re-upload them, then check the live page. Tracing each file's origin takes the most time.

1. Pull the asset list. Export your active ASINs, then capture every image slot, every product video and every A+ module (basic and Premium A+, plus Brand Story if you use it). A spreadsheet with one row per asset works fine.
2. Find the source files. Match each row to the master file in your drive or DAM. If an agency or freelancer made it, ask them in writing whether any person in it was generated. This is how you establish "actual knowledge" either way.
3. Classify each asset. Use four labels: real person, synthetic person, no person, non-photorealistic. Add a fifth, "unsure," for files nobody can trace.
4. Decide on the unsure files. If no one can confirm that a realistic person was photographed with a signed release, either tag the asset or replace it. We usually replace, because an untraceable face carries other risks too, such as resemblance to a real person.
5. Tag the final export, not the working file. Write the keyword into the exact JPEG, PNG or video file you will upload. The next section shows how.
6. Verify the tag is in the file. Read the metadata back before upload. Thirty seconds per batch saves a full redo.
7. Re-upload to every slot and every marketplace. Replace the old file in each image slot and A+ module where it appears. Shared A+ content and EU or UK copies of a listing are easy to miss.
8. Check the live page. Look for Amazon's indicator after processing. Because Amazon applies it "where applicable," its absence does not prove a problem, but keep a screenshot either way.
9. Log it. Record the ASIN, slot, file name, classification, who confirmed it and the date. That log is your evidence if a question ever comes up.

If your catalog is large, start with the ASINs that drive most revenue and any listing whose lifestyle images were made in the past two years, since that is when AI models became common in listing work. Our A+ content guide lists the module types worth checking.

How do I add the contains-synthetic-performer keyword to a file?

Open the final export in an editor that writes XMP metadata and add contains-synthetic-performer as a keyword in dc:subject. ExifTool, a free command line program, does this in one line per file or folder. Adobe Bridge does it through its Keywords panel. In both cases, read the field back before uploading.

Option 1: ExifTool (Windows, Mac, Linux)

ExifTool lists dc:subject as XMP-dc:Subject and treats it as a list, so you add an item instead of overwriting existing keywords. Its documentation shows XMP read and write support for JPEG, PNG, WebP, MP4 and MOV.

• Add the tag to one file: exiftool -XMP-dc:Subject+=contains-synthetic-performer lifestyle-01.jpg
• Tag every JPEG in a folder and its subfolders: exiftool -r -ext jpg -XMP-dc:Subject+=contains-synthetic-performer ./to-tag
• Check the result: exiftool -XMP-dc:Subject lifestyle-01.jpg should print a line that includes contains-synthetic-performer.

By default ExifTool keeps your untouched file with "_original" added to its name. Upload the edited file, not the backup. Once you trust the process, the -overwrite_original option skips the backup copies. Use the plus-equals form shown above: a plain equals sign replaces any keywords already in the field.

One trap: ExifTool's -Keywords tag writes the older IPTC IIM keyword field, which is a different location. The IPTC Photo Metadata Standard maps its "Keywords" property to XMP dc:subject, but Amazon named dc:subject (XMP) specifically, so target that field directly.

Option 2: Adobe Bridge

• Select the final export files in the Content panel.
• Open the Keywords panel, type contains-synthetic-performer into the field at the bottom of the panel, and click New Keyword.
• Tick the box next to the new keyword so a check mark shows for the selected files. Alternatively, use File, then File Info, and enter it in the keywords field before clicking OK.
• Run the ExifTool check above on one file to confirm the keyword landed in XMP dc:subject.

Adobe's interface changes between versions, and we could not confirm from Adobe's documentation exactly which fields each Bridge version writes, which is why the read-back step is there. Other IPTC-aware tools such as Lightroom Classic or Photo Mechanic can work too; the test is the same.

What mistakes and risks should sellers watch for?

Most problems come from tags that never reach Amazon, missed placements, or loose records. Writing the keyword into the wrong field, compressing away the metadata, forgetting videos and A+ modules, or relying on a freelancer's word with nothing in writing will each leave you exposed while you believe the job is done.

• Writing "AI-generated" on the image instead. Amazon asked for metadata, not overlay text. Text badges on secondary images can also clash with Amazon's image rules. See our main image requirements guide for what is allowed where.
• Tagging the PSD, then exporting a clean JPEG. The tag must be in the uploaded file.
• Forgetting video. Product videos and on-screen presenters are covered. If you produce video ads, our Sponsored Brands video guide covers creative specs.
• Ignoring ad creative and Stores. Amazon's notice names listings and A+ content. It does not mention Sponsored Brands custom images or Brand Store pages. New York's law covers advertising generally, so if a synthetic person appears in ad creative shown in New York, plan for disclosure there too and ask your account team how Amazon handles it.
• Tagging real people. Over-tagging puts an AI label on an honest photo. Keep the tag for synthetic people only.
• No written record from creators. Add a clause to agency and freelancer briefs requiring them to declare any generated person. Cooley's summary of the law recommends the same contract step.
• Look-alike faces. A generated face that resembles a real person can raise right of publicity issues that the disclosure does not fix. Replace any asset where that is possible.

There is also a policy risk. Amazon has not described how it will treat sellers who skip the tag. Image and content issues can show up in Account Health or as listing suppressions, so treat this as a compliance task with an owner and a deadline, not a creative preference.

How should we brief AI creative from now on?

Choose before production whether an asset will show a generated person, and write that choice into the brief. Use AI freely for scenes, props and backgrounds. Use real, released models where trust matters most, such as apparel fit, skincare results and anything health related. When you do use a synthetic person, plan for the tag and Amazon's label.

Here is the approach we take with brands:

• Separate people from scenes. A real model shot on a plain background, then placed into an AI-generated room, avoids the tag entirely and keeps the person authentic.
• Keep AI people out of claims. A generated "customer" showing before and after results invites trust questions even with a label. Show the product, the data and real reviews instead.
• Add a provenance line to every file name or DAM record. Something as simple as "SRC: shoot 2026-03" or "SRC: AI person" makes the next audit fast.
• Consider extra IPTC fields. The IPTC standard has a Digital Source Type property for marking AI-generated media. Amazon has not asked for it, so it is optional, but it helps keep internal records consistent.
• Test before you scale. Upload one tagged image, check how the listing renders, and watch conversion on that ASIN before you roll out a catalog-wide refresh.

Amazon itself promotes AI image tools to sellers and advertisers, so this rule is not a signal to stop using AI. It is a signal to know, file by file, what you are publishing. If you want a second pair of eyes on a catalog refresh, our listing optimization team builds this check into image production.

Do California or other places have similar rules?

California has an AI disclosure law, but it targets AI companies, not sellers. The California AI Transparency Act became operative on August 2, 2026 and requires generative AI providers with more than one million monthly users to embed hidden disclosures in AI images, video and audio, and to offer a free detection tool.

According to a Morgan Lewis summary, later phases extend obligations to large online platforms, AI hosting platforms and capture device makers in 2027 and 2028. That could one day affect how marketplaces display provenance data, but today the duty sits with the AI provider, not with an Amazon seller uploading images. We found no other US state law that, like New York's, puts an ad disclosure duty on the brand itself.

For sellers, the practical effect of California's law is that files from large AI image tools may already carry provenance metadata. That does not replace Amazon's keyword. Amazon asked for one specific value in one specific field, so add it even if the file already contains other AI markers.

What is the timeline, and what should be on my checklist?

The New York law has applied since June 9, 2026, and Amazon's tagging rule since its July 22, 2026 notice, so there is no grace period left to wait for. Work through the checklist below now for live listings, and add the same steps to your upload process for new assets going forward.

Checklist

• Asset inventory built for every live ASIN, video and A+ module, in every marketplace
• Each asset classified: real person, synthetic person, no person, non-photorealistic
• Synthetic assets tagged in XMP dc:subject with contains-synthetic-performer
• Tag verified by reading metadata back from the final file
• Tagged files re-uploaded and old versions replaced
• Creator briefs and contracts updated to require disclosure of generated people
• Upload process updated so the tag check happens before every new asset goes live
• Ad creative and Brand Store reviewed for synthetic people, with questions sent to Amazon where its notice is silent

If you would like an outside review, a SellerVine audit can include a pass over your image and A+ library alongside the usual listing and account checks.

Sources and data notes

Facts in this article were checked against the sources below on October 24, 2026. Amazon changes programs and dates often, so confirm details in Seller Central or the Amazon Ads console before you act.

• CNBC: Amazon makes sellers label AI-generated people in images after NY law
• New York State Senate: General Business Law section 396-B
• Cooley: New York Enacts 'Synthetic Performer' Disclosure Law for Advertisements
• Skadden: Two Newly Enacted New York Laws Will Regulate Certain AI-Generated Images
• Hunton: New York Enacts Law Regulating the Use of AI Generated Synthetic Performers in Advertising
• GeekSeller: Amazon Introduces New Rules for AI-Generated Images (July 2026), reproducing Amazon's notice
• My Amazon Guy: Amazon AI Image Policy Brings Mandatory Labels for AI-Generated People
• Ecomclips: Amazon New AI Image Rule 2026
• beBOLD Digital: Amazon Now Requires You to Label AI-Generated People
• ExifTool: XMP Tag Names (XMP-dc Subject)
• ExifTool: FAQ (list-type tags and += syntax)
• ExifTool: Application documentation and supported file types
• Adobe: Create and manage keywords in Adobe Bridge
• IPTC: Photo Metadata Standard (Keywords and Digital Source Type)
• Morgan Lewis: New California AI Disclosure Rules Become Operative

Frequently Asked Questions