I expected marketplace review to care about whether an image was AI-generated. Watching a few hundred listings go through, that is not what triggers action.
What triggers action is a mismatch between the image and the item that ships. The generation method is close to irrelevant on its own.
The three mismatches that cause problems
The item does not look like that. Colour that reads differently, a finish that is glossier than reality, proportions that are off. This is the most common and the most damaging, because it produces returns as well as review flags.
The image implies something not included. Props in frame that a buyer reasonably reads as part of the package. A charging cable, a case, a second item in a set. If it appears in the primary image without a clear caption, expect people to think it ships.
The image implies a capability the item does not have. Anything that reads as a demonstration of function. Presentation is fine. Function is not.
Nothing on that list is about how the image was produced. All three would be problems with a photograph too.
What review systems seem to look at
From what I can observe, and this is observation rather than any published rule.
Consistency between the primary image and the other images in the listing. A primary image styled very differently from the rest draws attention.
Consistency between the image and the written specification. If the copy says matte and the image is glossy, one of them is wrong.
Return and complaint rate on the listing. A listing that generates complaints gets looked at, and then the images get looked at.
That last one matters most. Review is often reactive. The image does not get scrutinised until something else goes wrong.
The boundary I work to
AI handles background, lighting, scene and composition.
The physical item does not get modified. Colour, finish, proportion, markings and included contents all come from the real product.
Anything a buyer would count or measure comes from the real product.
Working inside that boundary, the generation method has never been the issue. Working outside it, no amount of photographic realism helps.
Labelling
Rules on labelling AI-generated content differ by jurisdiction and by platform, and they are changing. Check the current requirements for the markets you sell into, and check them again periodically rather than once.
Two things are safe to assume anywhere. Do not remove or alter labelling that a tool has embedded. And if a platform asks whether content is AI-generated, answer accurately.
Before publishing
- The physical item in the image matches what ships in colour, finish and proportion
- Nothing in frame reads as included unless it is
- No implied capability the item does not have
- Primary image is stylistically consistent with the rest of the listing
- Image agrees with the written specification
- Embedded labelling has not been stripped by the processing pipeline
- Current labelling rules for the target market have been checked
Sources
This is drawn from my own production experience and from observing listing outcomes. It is not legal advice and it does not reproduce any platform's internal review criteria, which are not published. Platform rules change, so check each platform's current published policy.
Top comments (0)