Prepare Your Product Feed for AI Shopping Questions

Make your product data answerable before adding a shopping agent. Google’s September 16 commerce update discusses AI performance insights, conversational product information, Business Agent features and checkout developments. The retailer announcement describes different capabilities with their own access conditions. It is not a promise that every merchant account has every feature.

The useful first project requires no new agent: make one product’s important facts consistent across its feed, landing page and checkout. An assistant can phrase a persuasive answer from incomplete data; better phrasing does not fix an incorrect delivery promise or unsupported compatibility claim.

Choose one product family and collect recurring pre-purchase questions from your own approved support material. Remove customer identifiers. For a reusable bottle, relevant questions might concern capacity, dimensions, cleaning instructions, materials and replacement parts.

Separate a product fact from a policy. Capacity belongs to a variant specification; delivery depends on destination and current service; returns depend on applicable terms. Treating all three as static marketing copy makes it easy for an agent to give an outdated answer.

Use a worksheet that connects each answer to its owner and source. These are audit columns, not invented Google feed attribute names.

Buyer question	Verified source	Owning system	Missing fact
Will it fit?	Current dimensions	Product specification	Measurement method
Which version is this?	Variant identifier	Catalog	Ambiguous page option
When will it arrive?	Current delivery calculation	Checkout	Destination-dependent
Can I return it?	Applicable policy	Policy page	Regional exception

Check the current Merchant Center documentation for supported attributes and eligibility. Do not put a plausible custom column into a feed and assume Google will recognize it.

Open the exact variant page, inspect its feed record and walk to the last review step in a safe checkout preview without purchasing. Compare identifier, availability, price, currency and selected options. Ask the responsible owner to resolve mismatches at the source, rather than covering them with a generic disclaimer.

For digital goods, add questions about file format, license scope, delivery method, compatibility and update entitlement. Keep membership access separate from a one-time download. These distinctions can reduce confusion even if your catalog never enters a particular shopping feature.

Provide only approved product facts and a redacted export. Ask for missing information before asking for sales language.

Audit this product record and page text: {{verified_product_data}}
Common buyer questions: {{anonymized_questions}}
Return question | supplied evidence | source location |
conflict or missing field | responsible owner.
Do not invent dimensions, compatibility, stock, delivery or
license rights. Separate static product facts from live policies.
Suggest natural wording only for answers the evidence supports.

An illustrative finding could be that a page calls a bundle editable while the specification lists only flattened images. The correct next step is checking the actual files and correcting the catalog, not generating a more confident answer.

After approved changes, verify that the feed is accepted in the platform’s real diagnostics and that the public page still matches it. Test a few representative customer questions with the approved data. Include an unanswered question to check that the workflow preserves uncertainty.

Use existing performance reports where available, but do not describe feed cleanup as a ranking guarantee. Track useful outcomes such as fewer incorrect answers, fewer policy contradictions and better-qualified visits using your actual analytics.

For the visual side, see the consistent product campaign guide. Your next action is a one-family evidence audit. Accurate catalog facts make every later search, support and shopping workflow easier to trust.

Reviewed on 22 October 2026. Primary-source reporting and editorial analysis; the exercises and example results are illustrative, not hands-on performance measurements. Product access, regional rollout and commercial terms can change.

• Google: Agentic commerce updates, September 16

01 Choose a tool: compare the fit. 02 Use a prompt: start with context. 03 Check the result: keep your judgment.