AI-Generated Images on Affiliate Sites: Where They Help, Where They Cost Trust, and What to Use Instead of a Fake Product Photo
Image generators solved the stock photo problem and created a new one. A buyer guide no longer needs the same smiling-cyclist photo as every other guide; it can have its own.
It can also have a convincing picture of a bike light that doesn't exist, sitting beside a button for one that does, and the reader who notices the mount is wrong has learned something about the page that the rest of it won't recover from.
This post covers Google's position on AI images (and why it isn't the risk), the one use that backfires and why it costs more than it saves, the uses that work and replace stock photography well, labelling and alt text, the Amazon image rule that AI changes nothing about, and a decision table for each image slot in a buyer guide.
The Amazon images post covers what you may show of the product itself; this is everything else on the page.
The short answer
Use AI images for scene-setting heroes, diagrams and concept illustrations, where they replace stock photos with something yours. Never use one as a picture of the product; product imagery comes from Amazon, as the rules require and as the reader expects.
Label photoreal images as illustrations, write alt text about what's in the picture, and keep the content credentials in the file. Google doesn't care how the image was made; the reader does, in the one slot where it matters.
Google's position
Google's guidance on AI-generated content, images included, is that it's judged like anything else: useful or not, relevant or not, made for people or for rankings. Image search ranks images on the page's quality, the alt text, the file's size and format, and whether the image is near the text it illustrates, none of which depend on the generator.
Where an image carries content credentials declaring it AI-made, Google may show a label; that's a disclosure to the user, not a ranking signal against the page. The AI content and Google post covers the text side; the image side is the same answer.
The one use that backfires
A generated image of the product. It looks right at a glance, which is the problem: the reader is looking closely, because deciding whether to buy is what they came to do. The generator gets the brand's mark almost right, puts the charging port on the wrong face, adds a strap that the real one lacks, renders a button the product doesn't have.
A buyer comparing the picture with the Amazon listing finds the differences in seconds, and the conclusion they draw isn't "the illustration is off"; it's "this site hasn't seen the product", which is true, and which the rest of the guide was carefully not saying.
It also collides with the method claim. A guide that says it compared specs and reviews is honest; a guide that shows a photo-like image of each product implies something it didn't do. The reviewing without the product post covers how to be honest about method; a fake product photo undoes it in one image.
The uses that work
- The hero image. A scene for the reader, not the product: a commuter at dusk, a kitchen counter, a desk by a window. This is where stock photos lived, and a generated scene is better than stock because it's specific to the guide and nobody else has it.
- Diagrams. How a beam cut-off works, the layers of a mattress, where a standing desk's motor sits. Generated diagrams need checking (generators invent labels), but a checked diagram explains a spec that three paragraphs can't.
- Concept illustrations for informational posts: "why 1,000 lumens is too many for the city" wants a picture of glare, not a product.
- Section dividers and category art for hubs, where the image is decoration and the alt text is empty.
- Author-page backgrounds, not author portraits. A generated face as "the author" is the text version's fake testing claim.
Labelling and alt text
- Label photoreal images ("Illustration" in the caption or credit). Diagrams and obvious artwork don't need it.
- Keep the content credentials the generator writes into the file (C2PA metadata). Stripping them is possible and pointless; leaving them is the honest default and what Google reads for its label.
- Alt text describes the picture, not the product or the keyword: "a cyclist on a wet city street at dusk with a front light on" rather than "best bike lights for commuting". A generated image follows the same alt text rules as any other.
- Compress and size them like any image: a generator's output is large, and an uncompressed 4MB hero undoes the speed work.
The Amazon rule it doesn't change
Amazon's Associates policies allow product images on your site only through Amazon's own means (the API and its links), not saved files, and an AI image of the product is not an exception to that rule; it's a separate problem beside it.
The product's picture on an affiliate page comes from Amazon or isn't there. Everything else on the page is yours to illustrate however you like.
By image slot
| Slot | AI image? | Use instead or as well |
|---|---|---|
| Hero at the top of a guide | Yes, a scene for the reader | A real photo if you have one; stock if you must |
| Each product's picture | No | Amazon's image through the API or link |
| Comparison table thumbnails | No | Amazon's images, or no thumbnails |
| Explaining a spec | Yes, a checked diagram | A hand-made diagram; a labelled screenshot of your own |
| Informational post illustration | Yes | Stock, which it replaces |
| Author portrait | No | A real photo or an initial, never a generated face |
| "We tested" photo | No, under any circumstances | Your own photo, or don't claim it |
| Hub and category art | Yes | Decorative; empty alt |
Doing it with AffSERP
Guides written by AffSERP's X Best Review tool take product images from Amazon's API, so the product slots are filled the compliant way and never with a generated look-alike.
The featured image for each guide is a generated scene for the reader, sized and compressed, with alt text written about the picture; it fills the hero slot where stock used to go, and it's the only slot the tool generates for.
Diagrams and informational-post illustrations are yours to add, and the table above is the rule the tool follows: product pictures from Amazon, everything else may be made.
FAQ
Does Google penalise AI-generated images?
No. Google's guidance treats AI images as it treats any image: they're fine where they help the page and judged on the same things (relevance, quality, alt text, loading). Google has said it may label AI images in results where the metadata declares them, which is a disclosure, not a demotion. The risk on an affiliate site isn't Google; it's the reader who sees a generated picture of a product beside a buy button and, rightly, stops trusting the page.
Can I use AI images for Amazon affiliate products?
You can generate a picture that looks like the product; you shouldn't. It's a product that doesn't exist next to a link to one that does, and the differences (a port in the wrong place, a logo that isn't quite the logo, a strap that isn't there) are the things a buyer is checking. Product imagery on an Amazon affiliate site comes through Amazon's API or links, which is also what the Associates rules permit; AI images belong in the slots that aren't the product.
Where do AI-generated images work on a blog?
In the slots where the image sets a scene or explains an idea rather than shows a thing: the hero image of a buyer guide (a commuter at dusk, not a specific light), a diagram of how a spec works (beam cut-off, a mattress's layers), a concept illustration for an informational post, and author-neutral decoration. They replace stock photos, which were also not photos of your product, and they do it with an image nobody else has.
Should I label AI images on my website?
Label them where a reader might reasonably take the image as a photograph of something real: a caption or a line in the image credit ('illustration', 'AI-generated illustration'). A diagram doesn't need it; a photoreal scene does. Keep the generator's content credentials in the file where it writes them, since Google and some browsers read them. Labelling costs nothing and is the difference between an illustration and a deception when a reader notices.
AffSERP guides fill product slots through Amazon's API and generate only the hero scene, sized, compressed and with alt text about the picture.
Start with free wordsKeep reading
