AI-Generated Listing Images: What Amazon, Etsy and eBay Now Hold Sellers To

Within days of each other in late October 2026, two marketplaces said the same thing about AI-generated listing images. Amazon, in a staff post on the Seller Forums: "you are responsible for all content on your listing, including anything generated by AI." Etsy, announcing a change to its photo rules: "Your listings must include at least one unedited photo of the actual item against a real background." eBay, meanwhile, offers AI-generated scenes inside its own listing tools.

Generated images are now ordinary listing work, and the rules have caught up with them in one direction. This is each marketplace's position in its own words, what happens on Amazon when a customer says the product did not match the picture, and how we build an image set with AI in it that survives the complaint.

Amazon's post, in the Create and Manage Listings category, starts from the case sellers keep describing: a customer says the product does not match the listing, the image that set the expectation was generated, and the listing is removed. Amazon's answer is that the source does not matter. "Whether you used Amazon's built-in AI tools, a third-party generator, or created images manually, the standard is the same": the images "must accurately represent the product the customer will receive."

That includes Amazon's own tools, the point most replies seized on. The most supported reply came from a seller who had used AI only for scenes, never for the product, and still felt exposed. That is a reported experience, not a ruling, but the instinct is right: a generated setting is a different risk from a generated product.

Two other Amazon rules sit beside this one. A photorealistic generated person has to be declared, which the AI-generated people tag covers, and the main image still has to be a true photograph on white under Amazon's image requirements. The new post adds what happens when an image misleads.

"When a customer reports that a product doesn't match the detail page, Amazon investigates." The post places it under Amazon's misleading detail page policies and lists three outcomes.

The listing is hidden "from search until a compliant image is submitted". The Fix tab within Manage Products shows the suppression.

The account records it, as "A Product Condition Customer Complaint or Product Authenticity Complaint" on the Account Health Dashboard, which is how policy warnings begin to add up.

An appeal may follow, "with documentation, including invoices and a plan of action". A plan of action for an image complaint has to name the root cause, which is the image process, and the fix, which is a new image and the check that stops the next one.

So a generated picture that flatters a product does not stay a creative problem. With the first complaint it becomes an account health problem.

Etsy went further. Its help page says "Etsy requires at least one unedited photo for most items", and "An unedited photo should show the actual item against a real background and must not be edited or changed in any way." It asks that the first listing photo be "the clearest, most realistic representation of the item buyers will receive."

The other photos may use editing tools, "including AI, to adjust lighting or color, crop or resize images, or change backgrounds", "as long as those edits don't make the item appear different." Print-on-demand sellers may use a stock mockup its owner has authorised, and made-to-order items may show real finished items made before.

Enforcement starts gently. Etsy's staff post said buyer reports of AI misuse "won't trigger immediate enforcement" at first, and the help page says sellers will get tips when generated images are detected during listing. Digital Commerce 360 reported that Etsy published the guidelines on 22 October and that detection will reach existing listings over time; that part is reported, not on Etsy's own pages.

eBay's photo guidance in Seller Center supplies the tools. Sellers can "Choose from a variety of AI-generated scenes", use a background removal tool, and add "up to 24" photos at a minimum of 500 by 500 pixels, with 1600 by 1600 recommended. The same page sets the limit: "Try to only use this option for a few photos, and make sure the background doesn't distract from your item."

Its other rules are the same accuracy test in different words: "Don't add text, artwork, logos, or watermarks to your photos", "only show the variant that's being sold", and "Include closeups of blemishes, damage, or wear." On eBay the buyer's remedy for a misleading photo is a not as described case, so the photo set is also the seller's evidence.

TikTok Shop wrote its own version in September, with a label and a list of attributes AI may not alter, and TikTok Shop's AI content rules set them out.

Read together, the rules agree more than they differ. A generated setting is allowed, within limits. A generated product is not, wherever it changes what arrives: the colour, the finish, the size, the texture, the contents of the box. And they point towards at least one real photograph that anchors the rest, which Etsy now requires, Amazon's main image already demands, and eBay's advice implies.

In practice the edits that cause complaints are rarely dramatic. A tool that sharpens a fabric's weave, evens out a wood grain, lifts a colour towards the brand palette or fills a box with accessories sold separately has changed the product, though nobody asked it to. The safe edits change only what surrounds the product: the surface it stands on, the light, the room. That is the distinction a brief to a designer, or a prompt to a tool, has to state.

Our practice, for any of these marketplaces.

Photograph the product first. The main image and a true-colour detail come from a camera. Amazon's post tells sellers to "Keep your original product photos on file", and they are the evidence an appeal needs.

Generate the setting, never the product. The real product is placed into a generated scene. It is not regenerated, restyled or recoloured by the tool, and Amazon's own advice is to use AI "for lifestyle and secondary images".

Check it against the object. Amazon's forum staff advise comparing the AI output with the actual product, side by side, before publishing. We check colour, proportion, finish and what is in the box, and every scale reference against real measurements.

Version it per marketplace. The unedited photograph Etsy asks for, the variant-only rule on eBay, the white main image on Amazon.

Audit what is live. Amazon suggests "quarterly audits of older listings". We read the return and complaint reasons at the same time, because "not as described" is often an image promising what the product cannot deliver.

Why each shot in a set earns its slot, and why the camera still does most of the work, is the subject of eCommerce product photography that sells.

Planning image sets for each marketplace, with generated scenes where they help and photography where the rules or the buyer need it, is our listing images work. If a listing has drawn a complaint about its pictures, the image that set the expectation is where we would start.