Does AI image metadata survive my store’s CDN and export pipeline?

Map the delivery chain

Write down the creation method, editing application, exported master, uploaded asset and feed rendition. Keep the original generation record with the asset. Google’s current guidance requires appropriate IPTC DigitalSourceType information for AI-generated product images. Which value describes an image depends on how it was created; a partly edited photograph should not be randomly labeled as an entirely generated image.

A public seller question reports concerns about metadata and storefront processing. It is evidence that this question exists, not proof that every Shopify CDN removes the same fields or that the author’s rejection was correctly diagnosed. Inspect your own delivered files rather than adopting a comment as a technical finding.

Keep the image’s IPTC/XMP source declaration distinct from the digital_source_type used for generated title and description fields. They describe different submitted objects and use their documented formats. Do not copy the feed-text spelling into an image field merely because the names look similar. Retain how the image was actually made before choosing or repairing its provenance.

Compare bytes at each stage

Use a metadata inspector that can read the actual IPTC/XMP field. A filename containing “AI” or a generic EXIF date is not the same evidence. Preserve the inspection output privately alongside the versioned asset; do not publish unrelated location or device metadata.

For each stage, record the exact file or response version, byte length, content hash, inspector name/version and literal field result. A blank inspector result is ambiguous until you know that it reads that namespace and container format. Recheck with a supported inspection method before saying the host removed the field. Two equal-looking images may have different bytes; a changed hash alone does not prove which metadata changed.

Repair the last failing stage

For an illustrative generated perfume scene, the master can be correct while a resized derivative loses its provenance. Correct the export or delivery configuration at that stage, then inspect a fresh response. Do not repeatedly add a tag to the master if the next transformation discards it. Ask the hosting provider how to preserve the necessary field on the specific supported rendition.

Fictional sequence: master A reports the intended field, export B does too, while the feed response C lacks it under a verified inspector. Preserve those three inspection outputs and ask the provider about C’s transformation. If a corrected response D reports the field, record D’s version and recheck that the processed item points to D. Do not overwrite A and lose the original comparison.

Keep provenance and approval separate

A truthful source-type field does not prove that the bottle label, product variant or marketing claims are accurate. Review those against the real product too. This guide is a manual delivery review; CartVisual does not certify metadata on a third-party CDN. Record an unresolved field as unresolved and consult the destination’s current guidance before submitting the asset.

The seller discussion is retained as a historical question source; it was not independently reverified for this revision and supplies no current technical diagnosis. No actual merchant image or CDN transformation was tested here. For generated text, follow the separate structured copy field guide; for feed access failures, inspect delivery before inferring a metadata problem.