A product image is ready to upload, but the team is unsure whether an AI-edited person needs a label.

Fastest safe fix: decide whether the image contains a realistic person generated entirely by AI; if it does, add Amazon’s required keyword to XMP dc:subject before upload, then verify it in the final exported file.

Who this is for: Cross-border sellers preparing Amazon product images or A+ content.
Also for: Operators using macOS Preview and staff responsible for reviewing, uploading, and recording image versions.

This week: Check Amazon’s current guidance for your store and content placement before tagging a batch. Then test the workflow on a copy of one qualifying image and confirm the exported file retains the keyword.

Last updated October 4, 2026; requirements checked against Amazon’s metadata instructions, Seller Forums announcement, and Amazon’s linked image guidance.

Image owners: scope before tagging

The key distinction is not whether AI touched the file. It is whether the image includes a realistic person who was generated entirely by AI. Amazon’s guidance describes a metadata requirement for relevant AI-generated human imagery; it does not mean every AI-created background, product render, or retouched portrait automatically needs the same label. Start with the person depicted, not the editing tool.

Amazon’s specified keyword is AI-generated, placed in the image’s XMP dc:subject field. A filename such as ai-generated-model.jpg, visible wording printed on the image, or a value stored in a different metadata field is not a substitute. Follow Amazon’s metadata instructions for the required field and workflow.

For a real person whose existing photograph has been retouched with AI, do not treat “AI was used” as enough evidence that the person was fully generated. A background extension, skin cleanup, color correction, or removal of an object does not by itself make a real person an AI-generated person. If a face or body was replaced or substantially reconstructed, and your team cannot tell whether the result falls within Amazon’s description, pause the batch and check the current official guidance or ask Seller Support.

This cautious approach matters because an incorrect blanket rule creates two kinds of avoidable work: unnecessary metadata on images that may not be in scope, and missing metadata on images that are. Keep a short decision record for uncertain cases: source image, edits made, reviewer, and the guidance consulted.

What metadata do Amazon product images need?

For an image within the stated scope, add the exact keyword AI-generated to XMP dc:subject. Do not add it only to the filename, image caption, visible pixels, or another metadata field. Amazon’s guidance describes the metadata requirement; the IPTC XMP specification explains the XMP schema context for structured image information.

Operations teams: placement and policy boundaries

A metadata task and a listing-content review are separate checks. Before editing, identify where the asset will be used: a product detail page image, an A+ content image, or another placement. Amazon’s product image guidance and A+ content guidance address their respective content contexts. Check the latest rules for the target store and placement instead of assuming that one page settles every image location.

Amazon’s announcement says the requirement applies across its global stores. That is not a reason to extend the rule to asset types or upload locations that the current help pages do not clearly cover. When a placement is not explicitly addressed, record the uncertainty and verify it with Amazon before applying a broad team policy.

Do not confuse the image label with approval of the listing’s copy, product claims, or advertising. A correct metadata field does not establish that the content meets every image rule, policy, or legal requirement. Use the separate Amazon product image guidelines for image compliance, and review copy through the relevant listing process.

Before work begins, the operator should have a clear handoff:

  • Asset owner: identifies the image source and whether the person is fully AI-generated.
  • Mac operator: writes the keyword to the required XMP field on a working copy.
  • Reviewer: verifies the final exported file, not just the open editing document.
  • Seller Central publisher: uploads that verified version and records its destination.

This division prevents a frequent handoff gap: one person sees a keyword in an editor and assumes the file sent to the listing team contains the same metadata.

macOS operators: Preview workflow and field checks

macOS Preview can add image keywords. Use it for a controlled single-file workflow, and make a copy before editing so you can preserve the received original. Apple’s Preview instructions for adding keywords describe the available controls; interface details can vary with macOS version and file type.

How do you add the AI-person label in macOS Preview?

  1. Make a working copy. Duplicate the source image in your approved working folder. Keep the original unchanged and give the copy a name your team can distinguish.
  2. Open the copy in Preview. Confirm that you have opened the intended image, not an earlier export or a similarly named file.
  3. Open the image information panel. In Preview, show the Inspector and locate the image information and keyword controls. Apple’s guide to viewing image information explains how to inspect image details.
  4. Add the exact keyword. Enter AI-generated in the keywords control. Avoid adding explanatory text or a different spelling in place of the required keyword.
  5. Save the file. Save the edited copy, then close and reopen it. This gives the reviewer a chance to verify what was written to disk rather than relying on an unsaved editor view.
  6. Check the field and hand off the saved file. Confirm that the keyword is present in XMP dc:subject. Record the file version and pass that exact file to the reviewer.

Preview’s visible keyword control is a way to edit image keywords; it is not, by itself, proof that a later export retained the value in the required XMP field. If your workflow converts formats, resizes images, or sends assets through another editor, test that final route before applying it to a batch. The IPTC XMP specification describes the field structure, while Amazon’s instructions determine the required placement of the keyword.

How can you tell whether an AI-edited real photo is in scope?

Ask what happened to the depicted person, not merely what tool was used. If the underlying person is real and AI changed only the background or made routine retouching edits, that alone does not establish that the person was generated entirely by AI. If the edit replaced or generated the person, or the source and edit history cannot resolve the question, escalate it instead of guessing. Keep the decision tied to Amazon’s latest wording, not an informal team shortcut.

Reviewers and publishers: final-file evidence

The file that matters is the one the publisher uploads. A keyword shown in Preview can disappear if an image is flattened, converted, optimized, or exported through another tool. Compare the original, the processed copy, and the actual upload file. Confirm the exact keyword is still in dc:subject on the final version.

Use the image information panel as an initial check, then use a metadata-capable inspection method where needed to confirm the XMP field itself. Apple documents how Preview displays image information, but that display should not be mistaken for a guarantee about a separately exported file. When the final file’s field cannot be verified, stop the handoff and create a clean, checked export rather than relying on the editor’s earlier state.

A compact handoff record can include:

  • The asset filename and version identifier.
  • The product or content placement and target store.
  • The person-in-image scope decision and who made it.
  • Whether AI-generated was found in XMP dc:subject in the upload file.
  • The uploader and upload date, recorded using your team’s normal process.

Amazon may change its interface, guidance, or buyer-facing presentation. Do not promise that adding the metadata will create a visible customer notice, set a particular Seller Central status, or cause an image to pass review. Confirm any displayed status in the current Seller Central workflow and rely on Amazon’s latest official documentation for platform behavior.

Team leads: milestones, exceptions, and environment choice

Treat this as a controlled asset handoff rather than an editing shortcut. At intake, assign the scope decision to a named owner. At editing, tell the operator which copy is approved for modification. At review, require evidence from the final exported file. At upload, make the publisher responsible for matching the approved version to the destination. Apply the same checks when replacing an existing image; an earlier review does not verify a later replacement.

Pause bulk processing when the image’s person is ambiguous, the edit history is missing, the destination is not covered clearly by the current guidance, or the final XMP field cannot be inspected. Check Amazon’s current help pages or contact platform support before resuming. Amazon’s Seller Forums announcement is useful for tracking the announcement, but an unanswered seller question should not be treated as an official ruling.

Workflow option Best fit Main trade-off Decision evidence
Existing shared workstation Occasional edits when the team already has a controlled Mac process Files, editor state, and responsibility can be harder to track across operators Named owner, working-copy rule, and final-file field check
A team-owned Mac Frequent work with a stable, locally managed editing setup Your team is responsible for hardware availability, maintenance, and access Device access policy and documented export route
Remote Mac rental Temporary campaigns or a team that needs a macOS workspace without buying another Mac Remote access and file handoff must fit your network and security practices A trial run using a non-sensitive asset and the actual upload workflow

If the current process depends on a shared machine, a Windows-only workstation, or files passed between personal devices, the real drawbacks are inconsistent editing environments, unclear version ownership, and weak evidence of what was uploaded. A Mac can make the Preview-based workflow available, but buying hardware makes most sense when the workload is steady and your team wants to own and maintain the device. For a campaign, a temporary macOS task, or an evaluation before committing to hardware, a remote Mac may be a better fit: you can test the real editing and handoff process without treating the machine as an Amazon compliance shortcut.

You can review VMSPIN’s U.S. remote Mac option if your workflow calls for a macOS workspace, then compare VMSPIN’s pricing options against your expected use. Use a non-sensitive test image first, confirm the file path and access method meet your requirements, and keep the same scope and metadata checks regardless of where Preview runs. The label remains a file-level requirement; the remote environment does not guarantee listing approval, account status, or a buyer-visible disclosure.