Google Merchant Center AI Image Metadata: The Tag Sellers Strip

You generate a clean, on-brand AI product image, drop it into your Google Shopping feed, and it looks perfect. Then you crop it, resize it for web, or run it through a background remover, and something invisible happens: the file quietly loses the one piece of data Google now asks every AI image to carry. The picture is identical. The Google Merchant Center AI image metadata that was embedded inside it is gone. Most sellers never notice, because nothing on screen changes and no red error pops up.

This is the gap almost nobody is talking about. The AI disclosure conversation has focused on Amazon and Etsy, but Google quietly put its own requirement into the Merchant Center guidelines, and it lives inside the image file rather than in a form field. If you use AI to create or edit product photos, this is the compliance detail most likely to slip through your hands without you ever seeing it happen.

What Google Merchant Center now expects from AI images

Google's product data guidance is explicit about generative AI. When an image in your feed was created or edited with a generative AI tool, the file is expected to keep the IPTC DigitalSourceType property that identifies it as AI-originated. Google's instruction is blunt: Don't remove embedded metadata tags such as that property from AI-generated images. The point is provenance. Google wants to know, at the file level, whether a shopper is looking at a camera photograph or a synthetic render, and it wants that signal to survive from your generator all the way into the feed.

The requirement is not limited to your main photo. It applies across the image fields Google reads from your product data, including image_link, additional_image_link and lifestyle_image_link. So a gallery of AI-assisted lifestyle scenes is squarely in scope, not just the hero shot. There is a matching signal for text too: AI-generated titles and descriptions can be declared through the structured digital_source_type attribute, set to trained_algorithmic_media. Images carry the tag inside the file; text carries it in the feed.

The three values, and what each one means

The DigitalSourceType property is not a single on/off flag. It has three values, and they describe how the image came to exist:

• TrainedAlgorithmicMedia is the common one for product sellers. It marks an image created using a model trained on sampled content, which is exactly what most AI product photography tools produce.
• CompositeSynthetic covers a composite that includes synthetic elements, for example your real product photo dropped into an AI-generated background or scene.
• AlgorithmicMedia is for an image created purely by an algorithm with no sampled training data behind it, which is rare in ecommerce.

For a typical AI product image or an AI lifestyle scene built around your real product, the value that belongs in the file is one of the first two. The tool that generates the image should be writing it for you, because retrofitting it by hand across a whole catalog is painful and easy to forget.

Why the Google Merchant Center AI image metadata keeps disappearing

Here is the part that trips people up. Embedded IPTC metadata is fragile. A huge amount of everyday image handling strips it out silently, and the visible pixels never change, so there is no warning. The compliant image you generated on Monday can be non-compliant by Tuesday afternoon without a single deliberate act on your part.

The usual culprits are ordinary: exporting through "save for web" settings that discard metadata, resizing in a lightweight editor, screenshotting instead of downloading, running the image through a background remover or compressor that only keeps the pixels, or an upload pipeline that re-encodes files on the way in. Any one of these can drop the tag. Because the Google Merchant Center AI image metadata sits invisibly inside the file, the only way to know it survived is to actually inspect the file, which almost nobody does as part of their listing routine.

What is actually at stake

It helps to be precise here rather than alarmist. Stripping the tag does not throw an obvious rejection the way a wrong image size does. What it does is put you on the wrong side of a stated Google requirement, and it removes a provenance signal at exactly the moment platforms are leaning harder on provenance. Google is asking for the tag on purpose, and the direction of travel across every major surface is toward machine-readable proof of how content was made. Failing that quietly, catalog-wide, is not a risk you want sitting in your feed indefinitely, especially when the fix costs you nothing once your pipeline is set up correctly.

This is also not a Google-only story. The same pressure shows up on Amazon, where a photorealistic AI person has to be tagged in the file's metadata under the Amazon synthetic performer rule, and where broader AI image disclosure expectations now apply to listings. Google's feed requirement is the Shopping-side version of the same idea: keep the provenance data attached to the file. Treating it as one cross-platform habit, rather than a pile of separate rules, is what keeps it manageable.

How to keep your Google Merchant Center AI image metadata intact

The good news is that this is a solved problem once you stop fighting your own workflow. A short, practical checklist covers almost every case:

• Generate with a tool that writes the tag for you. If provenance metadata is embedded at creation, you are compliant from the first byte and only have to protect it after that.
• Stop re-exporting through metadata-stripping steps. The download that comes straight out of your generator is the safe copy. The more editors and compressors it passes through afterward, the more chances the tag has to disappear.
• If you must edit, use tools that preserve metadata. Some editors keep IPTC data on export and some discard it by default. Check the export setting before you push a whole catalog through it.
• Verify before you trust it. A quick metadata inspector, or a command-line tool like exiftool, will show you whether DigitalSourceType is still present. Spot-check a few images after any bulk process.
• Declare AI text through the feed. For AI-written titles and descriptions, use structured_title and structured_description with digital_source_type set to trained_algorithmic_media, so your text provenance is as clean as your image provenance.

Where this fits with the rest of your feed

Provenance metadata is one layer of a healthy Google Shopping feed, not the whole thing. It sits alongside clean product data and error-free syndication, and the same silent-failure pattern shows up elsewhere in the pipeline. If a metadata tag can vanish without warning, so can other feed attributes, which is the exact problem behind feed errors that quietly block AI shopping surfaces. The habit that protects you is the same one in both cases: treat the feed as something to inspect, not something to assume is fine because it looked fine when you set it up.

How ListingRVA keeps the tag attached

The cleanest way to stay compliant is to never let the metadata go missing in the first place. Our AI product photography embeds provenance metadata into generated images automatically, the same way it already tags photorealistic AI people for Amazon, so the file you download is compliant before you do anything else with it. And because the product in every image stays your real product, you are not trading provenance for accuracy: the picture represents what the shopper will actually receive.

If you sell on Amazon as well as Google, you can generate compliant Amazon product photography from the same real photo, keep the metadata intact across both feeds, and stop worrying about which platform wants which tag. One generation step, provenance built in, and nothing to strip out later.

Google Merchant Center AI image metadata FAQ

Does Google Merchant Center require metadata on AI-generated images?

Google's product data guidance says you should not remove the embedded IPTC DigitalSourceType property from images created with generative AI tools. In practice that means AI-generated or AI-edited product images in your feed are expected to carry that provenance tag.

What is the DigitalSourceType property?

It is an IPTC metadata field that records how an image was made. Its values include TrainedAlgorithmicMedia (made by a model trained on sampled content), CompositeSynthetic (a composite with synthetic elements, such as a real product on an AI background) and AlgorithmicMedia (created purely by an algorithm with no training data).

Which image fields does it apply to?

It applies to the images Google reads from your product data, including image_link, additional_image_link and lifestyle_image_link, so both your main image and your gallery images are in scope.

Does editing an AI image remove the metadata?

Often, yes. Resizing, cropping, "save for web" exports, background removers and compressors can all strip embedded IPTC metadata while leaving the pixels unchanged. Always verify the tag survived after any bulk edit.

How do I check if my image still has the tag?

Use a metadata inspector or a tool like exiftool to read the file. If DigitalSourceType is present, the provenance signal survived. If it is missing, the image was stripped somewhere in your workflow.

Does this apply to AI-written titles and descriptions too?

Yes. For AI-generated text, use the structured_title and structured_description attributes with digital_source_type set to trained_algorithmic_media, which is the feed-level equivalent of the image metadata tag.