Google Shopping Image Requirements (2026)

Merchant Center rejects images for reasons that are not obvious from looking at the image, and the reason is usually in a rule you did not know existed. Below is the current spec from Google's image link documentation, fetched September 10, 2026 — including one requirement that affects anyone using an AI tool for product photos and that almost no guide covers.

The dimensions, and the date that matters

Google has announced new image size requirements of at least 500 x 500 pixels for all products, with enforcement beginning January 31, 2027. Until then, the older, lower thresholds still apply, which is why you may currently have products serving happily at sizes that will stop qualifying.

The rest of the size spec:

Two specific traps in there. First, the file extension has to correspond to the actual format — Google calls out that naming a PNG as .jpg produces an "Unsupported image type" error, and that images synced from third-party platforms are a common source of this. Second, Google says explicitly not to scale up an image or submit a thumbnail. Upscaling a 400 px image to 500 px passes the dimension check and still looks like a 400 px image, which is why their recommendation sits at 1500 px rather than the minimum.

The AI metadata rule

This is the requirement worth reading twice, and it is stated in Google's image guidelines:

All images created using generative AI must contain meta data indicating that the image was AI-generated.

Google names the IPTC DigitalSourceType property and its values — TrainedAlgorithmicMedia for an image made by a model trained on sampled content, CompositeSynthetic for a composite containing synthetic elements, and AlgorithmicMedia for purely algorithmic output. It also says not to remove those embedded tags from images created with generative AI tools, naming its own Product Studio as an example of a tool that writes them.

One nuance worth getting right, from IPTC's own vocabulary: trainedAlgorithmicMedia is defined as media "created algorithmically using an Artificial Intelligence model trained on captured content" — its display name is now "Created using Generative AI". A photograph of your real product with a generated background is not that. It is closer to compositeSynthetic, defined as a "composite including generative AI elements," which Google also lists. There is a fourth code, compositeWithTrainedAlgorithmicMedia ("Edited using Generative AI"), that fits an inpainted or outpainted photo and that Google's page does not name. If you follow the shoot-the-product, generate-the-background workflow below, compositeSynthetic is the honest tag.

If you are generating product images with any AI tool and feeding them into a Merchant Center feed, that is now part of your workflow rather than a detail. Two practical consequences:

• Check whether your tool writes the tag. Many do not. Google's rule does not care whose fault it is.
• Check whether your pipeline strips it. Image CDNs and resize steps routinely drop metadata on re-encode. An image that left the generator correctly tagged can arrive at your feed untagged.

What gets a product disapproved

Google separates hard requirements from best practices. These are the requirements — break one and the product is disapproved and flagged in Diagnostics:

• No placeholder or any image that does not show your product.
• No generic image, graphic, or illustration in place of the actual product. Exceptions exist for Hardware (632), Vehicles and Parts (888), and Software > Computer Software (313).
• No logo or icon instead of a product image, except for Software > Computer Software (313).
• No single-color square. Exceptions: Vehicle Paint (3812), Craft Paint, Ink and Glaze (505378), Painting Consumables (503740).
• No promotional elements or overlays. Google's list is long and specific: calls to action, service information such as extended warranty or free shipping, price information, promotional adjectives such as "best" or "cheap", condition or compatibility text such as "new" or "2-piece", watermarks, brand or manufacturer or retailer names and logos unless inherent to the product itself, and barcodes.
• No borders.
• The image must accurately display the entire product with minimal or no staging. For packaging, show it multidimensionally rather than just the front of the box.
• For a bundle, show everything included. For a multipack, the main image shows a single unit.

The overlay rule is stricter than most sellers assume. A brand name printed on the actual product is fine because it is part of the product; the same brand name added as a corner graphic is not.

Best practices that quietly decide performance

• Solid white or transparent background for most cases. Staged and lifestyle images are acceptable if the product is clearly visible. Google warns not to use transparency for light-colored products, because they can render against black.
• Apparel worn by people. Avoid cropping the model's head or feet in full-body shots. Shoes, handbags, and accessories go alone in the main image and on a model in additional_image_link.
• One unique image per variant. A green couch and a blue couch need their own images, and a color variant image should show only that one variant.
• Submit only one image_link value. Extra entries are ignored; everything else goes in additional_image_link.

URL rules people get wrong

The image_link value is a URL of 1-2,000 characters, ASCII only, RFC 3986 compliant, starting with http or https. Symbols and spaces need encoding — an ampersand becomes %26, a comma becomes %2C. Your robots.txt has to let Googlebot and Googlebot-image reach the file, and changes there take 24-48 hours to take effect.

The recrawl behavior is the part that costs people weeks. Submit a new, unique URL and Google typically recrawls within about 3 days, showing the product as Under Review while the previous image keeps serving. Change the image content but keep the same URL and Google says it can take up to 6 weeks to notice. If you are refreshing your catalog photography, version the filenames.

Making the images

The realistic workflow for a small catalog is a real photograph of each product plus a generated background, not a fully generated product — Google's requirement that the image accurately display the entire product means a synthesized product is a disapproval waiting to happen, quite apart from being a returns problem.

The visual templates that fit the spec are Clean White Product Shot and Amazon White Background, both of which hold your product's shape, color, and label from a reference photo and replace only the surface and lighting. Image generation starts at 1 credit per image depending on the model, and new accounts get 10 starter credits once. If you are starting from existing catalog photos with cluttered backgrounds, changing a product photo background walks through that specific job.

Then export at 1500 px or larger, check the extension matches the format, add the AI metadata tag, and version the URL.

The same catalog usually has to satisfy several specs at once. Amazon wants pure white and 85% frame fill, eBay bans watermarks outright and forbids stock photos on used items, and Shopify leaves it to you. The intersection of all of them is a well-lit white background packshot at 1500 px or more, which is why that is the format worth shooting first.

FAQ

What size? At least 500 x 500 px, enforced from January 31, 2027; around 1500 x 1500 recommended; 64 megapixel and 16 MB ceilings.

Do AI images need labeling? Yes — Google requires metadata indicating AI generation, such as IPTC DigitalSourceType TrainedAlgorithmicMedia, and says not to remove it.

How much of the frame? 75-90%.

How fast does Google pick up a new image? About 3 days for a new URL; up to 6 weeks if you reuse the same URL with different content.