Amazon's AI Image Labeling Rule: What to Tag in 2026

Amazon now requires sellers to tag photorealistic, AI-generated people in product images and A+ Content with a specific metadata keyword: contains-synthetic-performer. It follows New York's new synthetic-performer disclosure law, and it applies globally across Amazon's stores, not just to New York shoppers.

That's the whole rule in one sentence. The rest of this post is what actually counts as a synthetic performer, what's exempt, exactly how to add the tag, and the honest gaps in what's confirmed so far.

What Actually Changed

Amazon told sellers in July 2026 to start tagging any product image or video containing a photorealistic AI-generated person. The tag goes into the file's metadata before you upload it, not into the listing copy or a checkbox in Seller Central. You add the keyword contains-synthetic-performer to the dc:subject field in the file's XMP metadata using an IPTC-compatible editor, then upload the tagged file as normal.

Once the metadata is there, Amazon reads it and, where applicable, shows shoppers a disclosure indicator on the listing so they know the image includes an AI-generated person. The mechanic is metadata-based, not a manual Seller Central toggle. If your image editing workflow doesn't preserve embedded metadata through export and compression, the tag can get stripped before it ever reaches Amazon, so that's worth checking before you assume you're covered.

This isn't Amazon inventing a compliance requirement out of nowhere. It's a direct response to a new state law, and understanding that law is the fastest way to understand exactly what Amazon is asking for.

The New York Law Behind This

New York General Business Law section 396-b, signed as S.8420-A/A.8887-B, took effect on June 9, 2026. It's the first synthetic-performer disclosure law in the country, and it requires advertisers to disclose when an ad features a "synthetic performer," a digitally created image or video built with generative AI or similar software to create the impression of a human performance, where that performance is by someone not recognizable as any identifiable real person.

Read that definition carefully, because the whole rule hinges on it. A synthetic performer isn't just "an AI tool touched this image somewhere." It's a fully fabricated person: nobody real posed for it, nobody real is being impersonated, the entire human being in frame was generated.

Penalties under the New York law are civil: $1,000 for a first violation, $5,000 for each violation after that. Those fines apply to the business running the ad under state law. That's a separate thing from anything Amazon might do on its own platform, and it's worth keeping those two consequences distinct instead of blurring them into one invented "Amazon fine." More on that in the honest-gaps section below.

The law also carves out two exemptions that don't really touch Amazon sellers directly but are worth knowing: expressive works like film, TV, streaming content, documentaries, and games are exempt where the synthetic performer is consistent with the underlying work, and audio-only ads are exempt outright. Neither applies to a standard product listing image, so don't lean on either as an out.

What Counts as a Synthetic Performer (and What's Exempt)

This is the part sellers get wrong most often, so here's the plain version.

In scope: a photorealistic person in your image or video who was generated entirely by AI and isn't based on any real, identifiable model. Think a "hero shot" lifestyle photo where the person holding your product doesn't exist. Think a spokesmodel-style image built by an AI generator instead of hired talent.

Not in scope, confirmed exemptions:

• A real model, photographed normally. Obviously fine.
• A real model whose photo was retouched or altered with AI tools. Still a real, identifiable person. Not a synthetic performer.
• Non-photorealistic characters: illustrated, cartoon, or video-game-style figures. The law and Amazon's rule both target photorealism specifically.
• AI-generated backgrounds, props, packaging mockups, or scenes with no people in them at all. The tag is about synthetic people, not synthetic imagery in general.

That last point trips up more sellers than any other part of this rule. A lot of AI image generation on Amazon right now is backgrounds and staged scenes, not people. None of that needs the tag. The rule is narrow: fabricated humans, not fabricated environments.

If you genuinely can't tell whether an image qualifies, the safer default is real photography. That sidesteps the question entirely and it's the one piece of advice in this post that requires zero judgment call.

How to Tag an Image: The Metadata Mechanic

Here's the actual mechanic, step by step.

1. Open the final export in an IPTC-compatible metadata editor. This has to happen after your last edit, right before upload, since re-exporting or recompressing a file can strip embedded metadata. Adobe Photoshop and Bridge write IPTC/XMP metadata natively. The free command-line tool ExifTool does too, if you're batch-processing a catalog.
2. Locate the dc:subject field inside the file's XMP metadata block. That's the specific field Amazon reads, not a generic keyword or caption field.
3. Add the exact keyword contains-synthetic-performer. Not a variation, not your own shorthand. Amazon's system is looking for that literal string.
4. Save with metadata preserved, and verify the tag survived the save (some export settings strip metadata by default to save file size).
5. Upload the tagged file to your listing images or A+ Content module as normal. No separate Seller Central checkbox exists for this. The metadata does the work.

For A+ Content specifically: the same tagging step applies before you upload into an A+ image module. There's no separate disclosure step inside A+ Content Manager itself, the tag lives in the file, not the module settings.

Does This Apply Outside New York

Yes, and this is the part most coverage undersells. The underlying law is a New York statute. Amazon's tagging requirement is not scoped to New York-facing listings. Amazon has said the rule applies across its stores worldwide.

That means a seller in Texas selling to a buyer in Oregon still needs the tag if the image qualifies. Amazon isn't running two catalogs, a tagged one for New York and an untagged one for everywhere else. Treat this as a global catalog policy triggered by a state law, not a rule you can geofence around.

The Checklist: Auditing Your Catalog

Here's the actual work, in order.

1. Inventory your A+ Content and listing images specifically for AI-generated people. Not AI-generated backgrounds or props, those aren't in scope. Just the images where a human figure was generated by AI.
2. For each one, confirm it's a fabricated person, not a real model. If a real person posed for the shot, even with AI-assisted retouching afterward, it's exempt. If nobody real exists in that frame, it needs the tag.
3. Apply the contains-synthetic-performer keyword to the dc:subject XMP field using a compatible editor, before you re-upload.
4. When you're not sure, default to real photography. It's exempt outright and removes the judgment call.
5. Check your export pipeline. If your workflow strips metadata during compression or a CDN re-encode, the tag needs to survive that step or it never reaches Amazon.
6. Extend the audit to video, not just static images. Amazon's requirement covers both. If you're running AI-generated spokesmodel video anywhere in your ad creative, it's in scope the same way a still image is.

Run this once across your active catalog, then make it a standing step in your image brief any time a designer or agency delivers new creative. It's a five-minute check per image once you know what you're looking for.

What the AI Shortcut Courses Never Mention

There's a whole cottage industry of listing-optimization courses teaching AI-generated "hero shot" images and synthetic spokesmodels as a hack: skip the photoshoot, skip the model fee, generate a person in Midjourney or a similar tool, and drop them into your lifestyle images. Cheaper, faster, no scheduling a shoot around anyone's availability.

None of those courses were talking about disclosure risk before this year, because there wasn't one. Now there is, and it's the kind of gap that doesn't show up until a state regulator or a competitor complaint puts it in front of you. The AI-generated-spokesmodel shortcut isn't dead. It's just no longer free of a compliance step, and most of the content teaching it hasn't caught up.

If you're using AI to generate people in your product imagery, the fix isn't to stop. It's to know which of your images are actually synthetic people versus real photography, before you touch the tag, not after someone flags a listing.

What's Still Unconfirmed

Being straight about the edges of this rule matters more than padding out a checklist. Here's what's genuinely not settled yet:

• Amazon hasn't published its own account-level penalty for missing the tag, separate from New York's civil fines. Don't assume a suspension or an account-health hit that hasn't been confirmed. What's confirmed is the tagging requirement itself and the state's fine schedule, which applies to the business, not to Amazon's own enforcement.
• There's no confirmed blanket deadline forcing you to re-tag every image already live in your catalog today. The conservative read: tag every new upload starting now, and re-tag any AI-people image the next time you touch it for an unrelated update, rather than treating this as a fire drill across your whole back catalog.
• Amazon hasn't detailed what happens on the shopper-facing side beyond "an indicator, where applicable." Exactly how prominent that disclosure is on the listing page isn't fully spelled out in what's public so far.

If you're building AI into your image pipeline anywhere in your catalog, our broader playbook on using AI on Amazon covers where AI tools genuinely help versus where a human still has to show up, and this labeling requirement is exactly the kind of guardrail that playbook was written around. If you're generating listing copy and image briefs with AI tools too, the ChatGPT prompts for Amazon sellers post has the compliance-pass step built into the workflow, worth running this new tag through the same review.

The Bottom Line

One tag, one field, one narrow definition. If an image in your catalog shows a fully AI-generated, photorealistic person, add contains-synthetic-performer to the dc:subject XMP field before you upload it. Everything else, real models, AI-altered real people, illustrated characters, AI backgrounds with no people, stays out of scope.

The bigger discipline this points to isn't the tag itself. It's knowing what's actually in your image library well enough to answer a compliance question in five minutes instead of scrambling through a full catalog audit after the fact. If you don't have a clear inventory of where AI touched your product imagery, that's worth fixing regardless of this specific rule.

Not sure what's actually sitting in your catalog, or want a second set of eyes on your listing and A+ compliance before this becomes a bigger cleanup? Run a free Amazon audit with us and we'll flag it alongside everything else. Or if you want a team handling this as part of ongoing account management instead of a one-time check, see how we run Amazon accounts.

Frequently asked questions