Product data readiness for AI answers

Run this before the AI citation investigation playbook finds a gap, not instead of it. Generative answer systems assemble buying answers from the feed, the rendered page, structured data, and whichever source they judge clearest, not from ranked links alone. Most of what determines whether a product is usable there is ordinary data hygiene, done ahead of time.

Owners

• Catalog: confirms identifiers, variant structure, and approved attribute values.
• Content: writes buyer-facing facts in plain, specific language.
• Storefront/dev: keeps structured data aligned with the rendered page.
• Discovery: runs the buying-question baseline and owns the retest.

Steps

1. Confirm GTIN, MPN, SKU, and canonical URL are unique and correct for each product and variant.
2. Complete Product and Offer markup: price, currency, availability, images, and variant relationships.
3. Write the buyer-facing facts a generic question would need (material, size or fit, compatibility, use case), not marketing copy.
4. Match those facts across the PDP, feed export, and channel diagnostic for the same product and market.
5. State shipping, returns, and merchant identity consistently across sources.
6. Check crawl access, robots rules, and indexability on the pages carrying that data.
7. Run a fixed set of buying questions against the product and record whether it appears and whether the facts are correct.
8. Log each gap as catalog, markup, content, or selection, and route it to the owning team.

Done when

• Identifiers, price, and availability match across catalog, PDP, structured data, and feed.
• Product and Offer markup validates against the rendered page.
• Buyer-facing facts are present, accurate, and free of unsupported claims.
• A baseline buying-question run exists to compare against after future changes.