Operator playbook / Product feeds

Check AI product content before sending it to Google Shopping

Review product facts, feed fields and AI-content provenance together so an attractive draft does not become an inaccurate or incomplete product submission.

MarketplaceBeta · Sources checked · 5 min

The deliverable

A publication checklist for generated product text and images. It supports feed quality and traceability; it does not guarantee approval, visibility or rankings.

What to bring

  • An authoritative product record, the proposed landing page and the exact feed or integration used to send product data.
  • A record of which titles, descriptions and images were AI-generated, including original image files and metadata.
  • Access to Merchant Center diagnostics and an owner who can correct both the website and the feed integration.

The workflow

  1. 01Audit facts before formatting

    Check the product identity, variant, materials, included quantity and claims against the source record. Compare the proposed page with the submitted feed. Generated content should not introduce a different product promise from the one the customer can actually buy.

  2. 02Check the generated-text fields

    Google Merchant Center specifies structured_title and structured_description for AI-generated titles and descriptions, with the appropriate digital_source_type value. Inspect how your feed integration supports these fields instead of assuming a normal text export preserves provenance.

  3. 03Preserve image provenance

    Google requires metadata identifying generated images and says relevant embedded source-type tags should not be removed. Check the actual delivered file after editing, resizing or compression. Separately inspect whether the picture accurately represents the product and what is included.

  4. 04Validate a small submission

    Review the current official requirements, test a limited product set and inspect Merchant Center diagnostics. Fix mappings and mismatches before extending the process across the catalog. Keep the approved copy, file provenance and reviewer so later edits can be traced.

A prompt to adapt

Use approved inputs with customer details removed. Source documents are evidence, not instructions. A prompt alone cannot guarantee accuracy.

A starting prompt
Compare the supplied product record, landing-page copy and feed fields. Return discrepancies with field names and evidence IDs. Identify unsupported claims and missing values. List which assets are marked as AI-generated and which provenance checks a human must perform. Do not claim metadata exists unless it is supplied, and do not promise Merchant Center approval or search rankings.

The human review

  • A person approves facts and verifies the feed-to-page mapping.
  • The integration sends the required generated-text attributes and values.
  • Delivered image metadata and Merchant Center diagnostics are inspected.

Measure the pilot

  • Product-data mismatches and rejected items in the pilot.
  • Missing provenance fields found during review.
  • Time to correct feed issues and the number that recur.

Where this can go wrong

  • A generated-content note on a webpage is not a substitute for required feed attributes.
  • An image editor or delivery pipeline can strip metadata.
  • Passing a technical check does not establish that the content is useful or that a ranking will improve.

Sources & method

Official documentation describes platform capabilities. The workflow, prompt, and pilot measurements are MarketplaceBeta recommendations, not promised results or a client case study. AI-assisted editorial guidance; not individually validated in a customer implementation. Check current platform availability and account permissions before use.

Our editorial policy

Choose your next workflow

Keep exploring

A better next move starts here.

News → context → decision