Google Merchant Image Requirements: the 75 to 90% fill rule, and the AI tag that breaks exports
Google publishes two things about product images that almost no seller knows: a fill range with an upper limit, and a mandatory metadata tag on every AI-generated image. Both are on the same help page, and both are routinely missed. Everything below is from Google Merchant Center Help, checked on 26 October 2026.
The specification, as published
| Requirement | Official value | What happens if you miss it |
|---|---|---|
| Minimum size | 500 x 500 px, for all products from 31 January 2027 | Disapproval once the date passes |
| Recommended size | 1,500 x 1,500 px or above | Reduced performance in listing formats |
| Maximum resolution | 64 megapixels | Rejected |
| Maximum file size | 16 MB | Rejected |
| Product fill | No less than 75%, no more than 90% | Disapproval |
| Borders | None | Disapproval |
| Formats | JPEG, WebP, PNG, GIF, BMP, TIFF | Unsupported image type error |
| Extension | Must match the actual format | A .jpg pointing at a PNG triggers an error |
| AI images | Must carry IPTC DigitalSourceType metadata | Policy violation, and the tag is often stripped by tools |
| Variants | A unique image per variant, one variant per image | Disapproval |
The upper bound that nobody quotes
Every article about product images repeats 85%, because that is Amazon's number. Google's number is a range: the product must take up no less than 75% and no more than 90% of the frame.
The floor is intuitive — too much empty space wastes the click. The ceiling is the interesting half. A tightly cropped image where the product fills 95% of the frame passes Amazon and fails Google, because Google's surfaces apply their own crops and overlays and need somewhere to put them.
Amazon's 85% sits inside Google's band, which is the only reason a single main image can serve both. If you crop to 95% because a tool told you 'fill the frame', you have optimised for one platform and broken another without any visible sign.
Google's AI rule is wider than Amazon's, and they do not agree
Amazon requires a tag on images containing photorealistic AI-generated people: the keyword contains-synthetic-performer in the XMP dc:subject field. Amazon is explicit that images with no people, or with real people, do not need it.
Google requires something different and broader: all images created using generative AI must contain metadata indicating that, specifically the IPTC DigitalSourceType property, with values such as TrainedAlgorithmicMedia for a generated image or CompositeSynthetic for a composite that includes synthetic elements. Google also states plainly: do not remove embedded metadata tags such as IPTC DigitalSourceType from images created with AI tools.
Line the two up and the same file can be compliant on one platform and non-compliant on the other. A generated background with no people in it needs Google's tag and does not need Amazon's. A generated model on a real background needs Amazon's tag and also Google's. The only safe policy is to tag every AI-touched image for Google and add Amazon's tag on top whenever a synthetic person appears.
The failure mode is mechanical rather than editorial: most product-photo tools and export pipelines rewrite or drop XMP on save. Google's warning about not removing the tag is really a warning about your export step, not your intent. Verify the metadata after export, on the actual file you upload.
What gets an image disapproved
- Any promotional overlay: calls to action, price or shipping text, warranty claims, adjectives like 'Best' or 'Top quality', condition labels like 'New'.
- Watermarks, copyright symbols, graphic stamps, or a brand name and logo added in post-production — allowed only if the logo is a physical part of the product itself.
- Barcodes, QR codes or serial tags overlaid on the photo.
- A border of any width.
- A placeholder or generic graphic: 'Coming soon', 'No image available', a store logo, a mockup or an illustration that is not the product.
- A single-colour image that is only a square of colour.
- Any product that is not sold together with the main product.
- One shared photo across multiple distinct products — Google requires a dedicated image per variant.
The date to put in the calendar
Google has announced that 500 x 500 px becomes mandatory for all products from 31 January 2027. Before that date, feeds with smaller images can still run; after it, they are disapproved.
The practical consequence is that a feed which has been quietly serving 400 px supplier images for years will fail all at once, on a date nobody is watching. If any part of your catalogue is sourced from supplier photography, audit the smallest images now rather than in January.
The exceptions are narrow and worth knowing: generic images, graphics and illustrations are allowed for Hardware, Vehicles & Parts, and Software categories, and single-colour images are allowed for vehicle paint, craft paint and painting consumables. Everything else is held to the same rules.
Sources
Every figure above comes from the pages below, read on the date shown. Marketplace policies change without announcement, so if you are about to rebuild a catalogue on one of these numbers, open the source and confirm it still says what it said.
- Google Merchant Center Help — Image link [image_link] — read Oct 26, 2026
This is one of 5 platform specs on the site. See how it lines up against the other 4 — including where they contradict each other — in the side-by-side comparison.
Tools that claim to handle this
The whole channel →Photoroom
Background removal and templated scenes, shaped for sellers rather than designers.
Flair AI
Staged product scenes with your product composited in, aimed at brand marketing.
Pebblely
Product photos on generated themed backgrounds, metered by credit.
Claid.ai
API-first image enhancement for marketplaces: cut out, upscale, relight, resize.
Pixc
Human-assisted product photo editing, sold per image.
Booth.ai
AI product photography tuned for furniture and large home goods.
Remove.bg
The cut-out everything else is compared against. Cut-out only.
Mokker AI
Drop a product onto generated backgrounds, with a plain-white preset.
None of these tools checks the numbers on this page for you. They produce the image; you still verify the corners, the fill percentage and the edge.