A+ content is where lifestyle photography lives on an Amazon listing: the product in a kitchen, on a desk, in somebody's hands. For two years the generative tools in A+ Content Manager have made that imagery cheap, and a good share of the people in those pictures were never in a studio. In July 2026 Amazon told sellers that those people now have to be declared.
This is the rule as Amazon wrote it, the four cases it excludes, the mechanics of the tag, what Amazon's forum staff have clarified since, the questions that are still open, and how to run A+ and listing imagery so the disclosure is a line item rather than a risk.
Yes, a photorealistic AI generated person in a listing image or video, or in A+ content, has to be tagged before it is uploaded, in every Amazon store. The tag is the keyword contains-synthetic-performer, added to the file's dc:subject field in its XMP metadata with an IPTC compatible metadata editor, and the A+ Content Manager also asks through its own checkboxes. When Amazon detects the tag, it shows shoppers that the image includes AI generated people. The sections below set out the rule, the tagging and what it means for a brand's creative.
Amazon's announcement to sellers is titled "You must now tag AI-generated people in listings and A+ content", and its reason is one sentence: "Recent legislation requires disclosure when images or videos in advertisements contain photorealistic AI-generated people." Amazon does not name the law; press reports connected the change to a New York statute on synthetic performers, and Amazon informed sellers on 22 July 2026 according to those reports.
The mechanism is a metadata keyword. Before uploading an image or video that contains a photorealistic AI-generated person, the seller adds the keyword contains-synthetic-performer to the dc:subject field of the file's XMP metadata, using an IPTC-compatible metadata editor. The scope is "product listing images, A+ content, and videos", and it applies in every Amazon store worldwide. When the tag is present, Amazon says it will show an indicator telling customers that the image includes AI-generated people.
The A+ Content Manager has its own version. Amazon's forum staff confirmed that the A+ upload interface carries checkboxes for marking content as "AI-generated" and for stating that it contains "AI-generated people". Sellers asked whether the checkbox alone satisfies the rule or whether the file metadata is still required; Amazon's post describes the metadata as the requirement, so the safe reading is both.
Amazon's post lists four exclusions, and they decide most real cases. The tag is not needed when the image contains only real people, even if those people were altered with AI. It is not needed for characters from films, television or games, or other expressive works. It is not needed for an image with no people in it. And it is not needed when the people are not photorealistic: an illustration, a cartoon, a stylised figure.
Amazon's staff added the practical corollary for retouching. AI used for background removal, colour correction or lighting is treated like conventional editing, provided the result is accurate and the image meets the standard image policies. So a real model, photographed in a studio and placed into a generated kitchen, is a real person and needs no tag. A generated model placed into a real kitchen does.
The rule is also prospective. Amazon's forum staff said existing published listings do not require retroactive disclosure, while recommending that older images are updated when a listing is next edited. That is the sensible route: an untagged AI model on an old A+ module is not a violation today, and it will be an untagged AI model forever unless somebody fixes it the next time the page is touched.
Amazon names no statute. The law the press reports pointed to is New York's S.8420, signed by Governor Hochul on 11 December 2025, which law firms summarising it say took effect 180 days later, on 9 June 2026. It applies to whoever produces or creates advertising that features a synthetic performer: a digitally created asset, made with generative AI or software algorithms, that is intended to create the impression of a human performer and is not recognisable as any identifiable real person.
Where the advertiser has actual knowledge that one is included, it must disclose it conspicuously. The law does not say in what words, where or at what size. A knowing violation carries a civil penalty of $1,000, and $5,000 for each one after. Advertising for films, television and games is exempt where the performer appears as it does in the work, and so are audio-only advertisements and AI used only to translate a real performer's words.
Read beside Amazon's post, the lines match: Amazon's exclusions for real people and for characters from expressive works follow the law's own. As summarised, the law also leaves out the media and platforms that only carry an advertisement, which on our reading is why Amazon asks the seller to declare and then shows the indicator itself. It also means the duty does not stop at Amazon. The same generated person in a brand's own social or search advertising is the brand's to disclose wherever the law applies. This is a summary of a statute, not legal advice, and a brand unsure about its own advertising should ask counsel.
The announcement drew more than forty replies and most of them asked the same things. Whether a partly visible person, a hand holding the product or a figure in the background, counts. How a seller can confirm the metadata survived Amazon's image processing. Whether the A+ checkbox is sufficient on its own. Amazon's staff acknowledged the questions and said answers would follow; at the time of writing the post carries no detailed clarification, and the help pages Amazon names ("How to tag media that contains an AI-generated person", the product image guide and the shoppable video guide) sit behind the Seller Central sign in.
The conservative answer to all three is the same: if a photorealistic generated person is visible at all, tag the file and tick the box, and keep the source files so the tag can be shown to have been applied. Nothing is lost by declaring a hand.
What This Changes About Lifestyle Content
Not as much as the replies suggest, and in one respect it is clarifying. The point of lifestyle imagery in A+ content is to answer the questions the bullets cannot: how big it is in a hand, how it looks on a shelf, who it is for.
What A+ and Premium A+ are each worth sets out which modules do that work. A generated person adds nothing to those answers that a real one does not, and now adds an indicator the shopper will read.
So the practical shift is where the people come from. A product photographed on a real person, with the background generated, carries no tag. A generated person carries one. For the hero modules, the comparison chart and the use-case blocks, a day of real photography still beats a generated set, for the reason
product photography that sells gives: the picture has to be true of the product the customer receives, and a generated hand holding a generated product is true of neither.
Where generated people are the only option, a long tail of products that will never justify a shoot, tag them, declare them in the A+ interface, and accept the indicator. The alternative, leaving the tag off, is a policy violation on a surface Amazon has just said it is watching, and
Amazon's policy warnings explains what a listing-level violation does to an account over time.
Most generated people now arrive from somebody else: a freelancer, an agency, a template pack, or a generative feature inside a design tool. The disclosure stays with the seller, and the New York law, as summarised above, turns on what the advertiser knew. So the brief has to make sure the seller always knows.
Ask for a declaration with every delivery: for each file, whether it contains a person, and whether that person was photographed or generated, in whole or in part. On our reading, a real model whose face has been replaced with a generated one is a generated person, since the face the shopper sees belongs to nobody, and a generated figure in the background is a person to declare on the conservative reading this article recommends. Ask for the tag to be in the delivered files, and for the source files and generation records to be kept for as long as the images are in use.
Put it in the contract as well as the brief. A clause that makes the supplier warrant the declaration, and replace an undeclared asset at its own cost, turns a rule the seller carries into a standard the supplier works to. The other half of the brief, what a generated scene may and may not change about the product itself, is a different rule, set out in
what Amazon, Etsy and eBay now require of AI-generated listing images.
The Main Image Is a Separate Rule
None of this loosens the main image. The main image must be a true photograph of the product on a pure white background, filling most of the frame, with no people in most categories and no text, and a generated lifestyle scene in that slot gets the listing suppressed regardless of any tag.
Amazon's 2026 image requirements cover that slot; the disclosure rule is about the secondary images, the A+ modules and the videos where people appear at all.
The field Amazon names is the one most image editors and asset libraries label Keywords. The IPTC photo metadata standard stores its Keywords in XMP as dc:subject, so entering contains-synthetic-performer in the Keywords of the file's information panel, spelled exactly and as a keyword of its own, normally writes it where Amazon looks. Most video formats can carry XMP as well, though fewer editors write it, so the exported video needs the same check.
The risk is losing it on the way. Many export and compression settings strip metadata to make a file smaller, and a picture saved for the web, run through a compressor or exported from a design file can arrive without the keyword the original had. So read the tag back from the final file, the one that will be uploaded, with a metadata viewer, not from the working file.
Make it the last step of the export rather than a job for whoever uploads. A folder of finished files named by ASIN and image slot, each one read back and marked in the asset register as tagged, means the person at the upload screen ticks the A+ boxes against a list rather than from memory. Whether the tag survives Amazon's own processing is one of the questions Amazon has not answered, which is the reason to keep the tagged originals: they are the evidence that the seller did its part.
Keep an asset register that records, for every image and video on every listing, whether it contains a person, whether that person is real or generated, and whether the file carries the tag. Add the metadata before upload, not after, because the keyword has to be in the file Amazon receives. Tick both A+ checkboxes where they apply. When a listing is edited for any reason, review its older modules against the register and fix what needs fixing. And prefer real people in the modules that sell, because the indicator is a sentence on the page and a shopper reads it.
That register, and the modules built to it, are part of what our
A+ content service produces: the comparison chart, the use-case blocks, the brand story and the lifestyle imagery, with every person in every picture accounted for. If your A+ pages were built with the generative tools before July 2026, a review of what is in them is a short job and worth doing before the next edit.