Are AI-Generated Images Allowed on Amazon? The 2026 Policy, Read Carefully

Sellers ask whether Amazon will ban their AI images. Amazon has never answered that question, because its image rules are not written about production methods — they are written about output. As of August 2026 there is no Amazon prohibition on generating a listing image, exactly one AI-specific disclosure rule, and a great deal of confident blog advice in between that nobody has sourced. This is the policy explainer: what Amazon documents, what is practitioner consensus, and where the risk that actually costs money sits.

What Amazon writes down

Amazon's image requirements are technical and representational. The main image must show the actual product on a pure white background (RGB 255,255,255), with the product filling roughly 85% of the frame, and no added text, logos, watermarks, borders or props — and no illustration standing in for a photograph. The longest side needs 1,000 px for zoom to work, with 1,600 px or more recommended.

Gallery slots two to nine are governed far more loosely. Text is allowed. Diagrams, comparison tables, dimension callouts and lifestyle scenes are all normal, which is why every infographic set lives there.

Nothing in any of that mentions how the file was produced. The substantive rule underneath is representational: the image must not show a colour, size, quantity, component or capability that the shipped item does not have. That rule pre-dates generative models by two decades, and it is the one they make easiest to break by accident.

The strongest evidence of Amazon's stance is a product, not a policy

If you want to know what a platform thinks about a technique, look at what it builds. Amazon launched AI Creative Studio on 15 October 2024, and it is free to use with an Amazon Ads account. It generates lifestyle imagery, video and audio from your existing product photograph. The earlier image generator in the ads console was released on the same terms — Amazon describes it as available at no additional cost.

How it works matters more than that it exists. You supply a real product photo; the system places it into a generated scene. That is precisely the shape the image rules permit: real product, invented surroundings. A marketplace does not ship a free tool that produces policy violations.

One caution about the numbers attached to it, because they get quoted badly. Amazon has published that lifestyle-context imagery in mobile Sponsored Brands ads can produce click-through rates around 40% higher than standard product images, and around 10.3% higher return on ad spend for Sponsored Brands campaigns using AI-generated images. Those are advertising figures, for ad placements. They are not measurements of what a generated gallery image does to conversion on a detail page. As far as we can find, Amazon has published no listing-image conversion figure for generated creative at all, and anyone citing the 40% as if it were one is repeating an ad statistic in the wrong room.

The one AI-specific rule that exists

On 22 July 2026 Amazon added a disclosure obligation, and it is narrower than most of the coverage implied. If a listing image, video or A+ Content module contains a photorealistic AI-generated person, you must write the keyword contains-synthetic-performer into the file's dc:subject XMP field using an IPTC-compatible metadata editor, before upload. Amazon then shows shoppers an indicator on the content.

What it excludes is most of what sellers do: real people edited or enhanced with AI, non-photorealistic characters, fictional characters, and images containing no people at all. There is no equivalent tag for a generated background, a generated surface, or a re-rendered product. If your set has no human faces or hands in it, this rule never touches you.

The trigger was a New York law requiring advertisers to disclose synthetic performers, carrying civil penalties of $1,000 for a first violation and $5,000 for subsequent ones. Amazon applied the requirement across its marketplace rather than to New York traffic alone.

A separate regime is routinely conflated with it. Article 50 of the EU AI Act obliges providers of generative systems to mark synthetic outputs in a machine-readable format, applying from 2 August 2026, with systems already on the market before that date given until 2 December 2026 under the May 2026 omnibus agreement. That is a duty on whoever builds the model, not a labelling requirement on your listing. Article 50 also places a disclosure duty on deployers of "deep fake" image content; whether a generated backdrop around a real product falls inside that definition is genuinely unsettled, and we have found no regulator or Amazon statement addressing marketplace listing images specifically. Treat anyone who tells you it is settled with suspicion.

Where enforcement actually comes from

There is no evidence Amazon runs an AI detector over listing images, and no published policy that would give it anything to act on if it did. Enforcement arrives the ordinary way: a customer receives something that does not match the picture, returns it under "not as described", writes the review, or files an A-to-z claim — and the listing then gets looked at under the accuracy rules that have always existed. The generated image is never the charge. The mismatch is.

Which makes the expensive failures predictable, and worth checking before upload rather than after. Colour drift on a variant. Scale that implies a larger object. An extra unit, cable or accessory that the box does not contain. Fabricated text or certification badges on packaging. Textures and fine print that a model has smoothed into something more premium than the item. Hands and fingers that betray the render. Every one of these is a returns driver first and a policy matter second, and returns are not cheap: US retail returns were roughly $849.9 billion in 2025 on National Retail Federation figures, near 15.8% of sales and around 19.3% online.

A practical rule for where generation is safe

Three questions, asked of any pixel you did not photograph.

Does this pixel make a claim about what arrives in the box? A wooden countertop does not. The number of sachets does. Generate the first freely; never the second.
Would the buyer notice the difference holding the product? If yes, that region needs to come from a real photograph, not a re-rendering of one.
Is it slot one? Then it is a photograph on pure white, with no exceptions, no composites and no lifestyle scene. The main image loses arguments the gallery wins easily.

Then one procedural step: if a photorealistic person appears anywhere in the set, tag the file before upload, or hire a real model and skip the obligation entirely.

Documented versus widely believed

Statement	Status
Main image must be a real photograph on pure white, ~85% fill	Documented by Amazon
Photorealistic AI people must carry `contains-synthetic-performer`	Documented, 22 July 2026
Amazon has no rule against AI-generated images as such	Documented by absence — the guidelines address output only
Generated backgrounds around a real product are acceptable	Consensus, strongly supported by Amazon shipping exactly that tool
Amazon detects AI images automatically	No evidence found either way
AI imagery helps or hurts listing conversion	No published figures; the 40% CTR number is about ads

The short version: Amazon does not care how you made the image. It cares, consistently and expensively, whether the buyer receives what the image promised. That standard is older than the technology, and it is the only one you need to design against.