Yes. Google requires every image created with generative AI and submitted to Merchant Center to contain embedded IPTC metadata identifying its digital source. Usually, that means retaining the DigitalSourceType value TrainedAlgorithmicMedia. A visible “AI-generated” label is not specified for Merchant Center images.
Which images need identification?
Google’s requirement applies to all images created using generative AI. These images may be submitted through any of the following Merchant Center attributes:
- Image link
[image_link] - Additional image link
[additional_image_link] - Lifestyle image link
[lifestyle_image_link]
The identification must be embedded in the image as an IPTC DigitalSourceType metadata tag. Adding “AI-generated” to the filename, feed description or image alt text is not a documented substitute for that metadata.
Google lists three source codes that should be preserved when applicable:
| IPTC source code | When it applies |
|---|---|
TrainedAlgorithmicMedia |
The image was created using a model derived from sampled content. This is the relevant code for generative-AI images. |
CompositeSynthetic |
The image is a composite containing synthetic elements. |
AlgorithmicMedia |
The image was created entirely by an algorithm that was not based on sampled training data, such as a mathematical formula. |
These definitions and the Merchant Center requirement are documented in Google’s AI-generated content guidance.
What should sellers check before submitting an image?
Confirm both how the image was produced and whether its embedded metadata survived your publishing workflow.
1. Establish whether generative AI created the image
Treat an image as generative-AI content when a generative model created it. For example, this includes a synthetic product scene generated from a prompt or product photo.
Do not choose the source code based only on how realistic the result looks. The code describes the image’s digital source, not its visual style.
Google’s guidance separately identifies composite and purely algorithmic images. Use those codes only when they accurately describe how the asset was made.
2. Find out whether the generating tool embedded the metadata
Check the exported image with software capable of reading IPTC metadata. Look for the DigitalSourceType property and verify that it contains the applicable source value.
If the image generator supplied the correct metadata, preserve it. Google specifically warns merchants not to remove embedded metadata tags from generative-AI images, including images made with Google’s Product Studio.
3. Inspect the final hosted file
Do not check only the original file on the designer’s computer. Inspect the exact image served at the URL placed in your Merchant Center feed.
Exporters, image optimizers, content-management systems and delivery workflows can create a new file. If a downstream process strips the embedded metadata, the original source file’s compliance does not fix the final hosted copy.
This final-file inspection is a practical merchant recommendation rather than a separate Google rule.
How should you prepare the image?
Use a preservation-focused workflow:
- Obtain the source information. Ask the person or tool that created the image how it was made and which digital-source code applies.
- Export the asset with metadata intact. Avoid settings that deliberately discard metadata.
- Inspect the exported file. Confirm that the relevant IPTC
DigitalSourceTypevalue remains embedded. - Process the image as usual. Resize or compress it only through a workflow that retains the metadata.
- Upload the image to its permanent location.
- Download or inspect the hosted version. Verify that the embedded source information is still present.
- Place that URL in the applicable attribute. Use
[image_link],[additional_image_link]or[lifestyle_image_link]as appropriate. - Keep the verified source asset. This gives you a known-good version if a later export or hosting change removes its metadata.
The central task is not adding a marketing caption. It is maintaining machine-readable provenance from creation through delivery.
Example: an AI-generated lifestyle image
Example — hypothetical Merchant Center workflow
A furniture merchant has a real product photograph of a blue chair. The merchant uses a generative model trained on sampled content to create a new image showing that chair in a synthetic living-room scene.
The resulting lifestyle image is generative-AI content. Its applicable IPTC source value is:
DigitalSourceType: TrainedAlgorithmicMedia
The merchant then:
- Exports the generated image without deleting its metadata.
- Confirms that
TrainedAlgorithmicMediaappears in the exported file. - Compresses the image using a workflow configured to preserve metadata.
- Uploads it to the store’s image server.
- Checks the image returned by the public URL.
- Submits that URL through
[lifestyle_image_link].
If compression strips the IPTC property, the merchant should not rely on the metadata present in the earlier file. The delivered image must be corrected or replaced before it is used.
The merchant should also review the image against the actual chair. The metadata identifies the image’s origin; it does not establish that the generated product details are accurate.
How should product details be handled?
Identification and product accuracy are separate responsibilities. An embedded AI source tag does not make an inaccurate product representation acceptable.
Before submitting the image, compare it with the actual item and approved product photographs. Check details shoppers may rely on, including:
- Color and pattern
- Shape and proportions
- Materials and finish
- Included parts
- Logos and lettering
- Variant-specific features
If the synthetic image changes a meaningful product feature, regenerate or replace it. Do not use the provenance tag as a disclaimer for a misleading product image.
What does an input-and-output comparison show?
The following is a supplied public example of an original product input and a completed AI-generated output. It illustrates the type of comparison a merchant can perform, but it does not prove universal fidelity or establish Merchant Center compliance.
Case example: Jeneva Lace Midi Dress — Brown
Original product input: View the original dress image

Completed output thumbnail: View the generated output thumbnail

Completed output video: Watch the generated dress video
For an image intended for Merchant Center, the operator should compare visible details such as the dress color, lace placement, neckline, sleeves and silhouette. The operator must separately inspect the actual image file for the applicable IPTC metadata. A visual comparison cannot reveal whether metadata is present.
This case is a curated public input-output pair, not evidence that every generated asset will preserve every detail.
Why might the metadata be missing?
Work through the asset chain in order.
Was it absent from the first export?
If the original generated file has no applicable IPTC source value, the problem began at creation or export. Confirm the image’s source and add or restore the accurate metadata through an appropriate metadata-capable workflow.
Do not guess between TrainedAlgorithmicMedia, CompositeSynthetic and AlgorithmicMedia. Use the value that reflects how the asset was actually produced.
Did an editing or compression step remove it?
Compare the metadata in:
- The original generated file
- The edited file
- The compressed file
- The final hosted file
The first version where the property disappears identifies the stage that needs correction. Re-export that stage without stripping metadata or use another preservation workflow.
Does the local file pass while the URL does not?
The hosting or image-delivery process may be returning a transformed copy. Inspect the exact public URL used in the feed rather than assuming it serves the uploaded original.
A practical fix is to change the delivery configuration or host a compliant copy through a path that preserves embedded metadata.
Is a visible disclosure enough?
No. A caption, watermark or filename may provide human-readable context, but Google’s documented Merchant Center requirement is the embedded IPTC DigitalSourceType metadata. Do not use a visible disclosure as a replacement for the machine-readable tag.
When does this image advice not apply?
This specific requirement concerns images created using generative AI in Merchant Center. It does not mean that every ordinary edit to a conventional product photograph automatically requires the TrainedAlgorithmicMedia value.
The supplied Google guidance does not define every possible boundary between generative creation and conventional image editing. If an asset’s production history is unclear, confirm it with the creator or software provider rather than deciding from appearance alone.
The guidance also distinguishes generative images from other kinds of synthetic assets. Preserve CompositeSynthetic or AlgorithmicMedia when those codes accurately describe the source instead of replacing them indiscriminately with TrainedAlgorithmicMedia.
FAQ
Do Merchant Center sellers need a visible AI label on the image?
No visible label is specified in Google’s Merchant Center guidance. Google requires generative-AI images to contain the applicable embedded IPTC DigitalSourceType metadata.
Other destinations may impose their own visible disclosure rules. Compliance in Merchant Center does not automatically satisfy requirements elsewhere.
Does the requirement apply only to the main product image?
No. Google says AI-generated images can be used in [image_link], [additional_image_link] and [lifestyle_image_link], and its metadata requirement applies to all images created using generative AI.
Which value should a typical generative-AI product image use?
Use TrainedAlgorithmicMedia when the image was created using a model derived from sampled content. That is the source type Google identifies for generative-AI images.
Can I remove metadata to reduce the image file size?
Do not remove the required source metadata. Google explicitly tells merchants not to remove embedded metadata tags such as the IPTC DigitalSourceType property from images created with generative AI.
Does Google Product Studio handle this automatically?
Google’s guidance says not to remove embedded metadata from generative-AI images and names Product Studio as an example. Even so, preserve the metadata through later exporting, optimization and hosting steps.
Are AI-generated titles and descriptions handled the same way?
No. Google documents separate feed fields for generative-AI text. AI-generated titles use [structured_title], and AI-generated descriptions use [structured_description], with the required [digital_source_type] sub-attribute set to trained_algorithmic_media. See Google’s AI-generated content documentation for the text requirements.
Does retaining the tag prove that the product image is accurate?
No. The tag identifies the digital source; it does not verify product fidelity. Review the generated image against the real product and approved listing materials before submitting it.
Alvin’s Club Studio is one option for creating AI product media; Merchant Center assets still need to follow Google’s metadata requirements.
Related on Alvin's Club Studio
About the author
- Founder building AI-native fashion commerce infrastructure. I design autonomous systems, agent workflows, and automation frameworks that replace manual retail operations. Currently focused on AI-driven commerce infrastructure, multi-agent systems, and scalable automation.
Comments
Join the discussionNo comments yet. Be the first to comment.