When a shopper asks ChatGPT "what are the best waterproof hiking boots under $200 for wide feet?", the AI doesn't browse your store the way a person does. It scans structured data and text for products that explicitly match four constraints: waterproof, hiking, under $200, wide fit. If your catalog doesn't state those attributes in a machine-readable way, your boots are invisible β filtered out before the shopper ever sees them.
This guide is about doing that at catalog scale. Classic ecommerce SEO optimized for keywords like "hiking boots." AI shopping optimizes for attributes and relationships, so an agent can confidently say "this is exactly what the user asked for." Below are five steps: maximize attribute density, add complete `Product`/`Offer` schema, structure product Q&A, keep your feed clean, and monitor which SKUs actually win the AI card.
Key takeaways
- AI agents filter products by explicit attributes, not marketing adjectives. "Comfortable winter jacket" is invisible to a shopper searching for "waterproof down parka, 700 fill, black." - Fill every relevant field. Push attribute completion toward 95%+ across material, color, size, fit, GTIN, and use case β the fields agents match against. - Structure your Q&A. Sizing, care, compatibility, and returns belong in `FAQPage` schema and clean on-page blocks, not buried in a paragraph, so agents can lift the answer. - Complete `Product`/`Offer` schema β including price, availability, `AggregateRating`, and `MerchantReturnPolicy` β is how the model retrieves accurate facts instead of hallucinating them. - GEOly's catalog and Share of Card analysis show which SKUs get an AI card, which attributes are missing, and where retailer quotes intercept your traffic β so you fix causes at scale.
Step 1: Maximize attribute density
AI models filter on specific attributes, not vague marketing language. A title like "comfortable winter jacket" tells an agent almost nothing; a shopper looking for a "waterproof down parka, 700 fill, black" will never match it. The gap between those two is attribute density, and closing it is the highest-leverage catalog change you can make.
Fill every field a shopping feed and structured-data spec offers. Be specific: material is "100% recycled polyester," not just "polyester"; pattern is "geometric," not "cool design"; occasion is "business casual, wedding guest." Populate size, color, fit, GTIN, brand, and use case on every SKU. Aim for 95%+ attribute completion across the catalog, and treat empty fields as lost recommendations.
Gotcha: specificity has to be accurate, not padded. Stuffing irrelevant attributes to look complete confuses matching and can misfile a product into queries it can't satisfy. Precise beats verbose.
Step 2: Implement product Q&A schema
Shoppers ask agents concrete questions β "can this be machine washed?", "does it fit wide feet?", "what's the return window?" β and the catalog that answers them explicitly wins the recommendation. Don't bury those details in a description paragraph; structure them so an agent can extract the answer directly.
Use `FAQPage` schema nested on the product page, with each entry a clean question and answer: Q "Is this machine washable?" A "Yes, cold cycle. Do not tumble dry." Cover the questions that actually drive purchase decisions in your category β sizing, care, compatibility, warranty, materials. Mirror the same content in a short on-page FAQ block so both humans and models see it.
Gotcha: keep the answers factual and current. An agent that lifts "machine washable" from your schema and a customer who ruins the garment is a trust and returns problem. Update the Q&A when the product spec changes.
Step 3: Complete Product and Offer schema
Structured data (Schema.org) is no longer just for rich snippets β it's the primary way an AI understands your products. Each product page should expose a complete `Product` object with a nested `Offer`, so the model retrieves facts instead of inventing them.
Fill the fields agents use to compare and recommend: `Offer` with `price`, `priceCurrency`, and `availability`; `AggregateRating` and `Review` with real numbers; `MerchantReturnPolicy` with the return window; `shippingDetails` with cost and delivery estimate; and product attributes like `material`, `color`, `size`, `gtin`, and `brand`. Many themes already emit partial schema, so the job is completing it, not duplicating it. Validate every page in Google's Rich Results Test before shipping. For the store-level walkthrough, see the [Shopify GEO optimization guide](/blog/shopify-geo-optimization-guide-geoly).
Gotcha: never invent ratings or policies to look stronger. Agents cross-check against reviews and third-party sources, and false schema erodes trust and can trigger penalties. Publish only what's true.
Step 4: Keep the feed clean and consistent
A catalog optimized page-by-page still loses if the underlying feed is inconsistent. The same SKU described one way on the product page, another in the Merchant Center feed, and a third on a marketplace gives the model conflicting facts β and conflict makes it hesitate or pick a cleaner competitor.
Align titles, attributes, prices, and availability across your product pages, structured data, and any shopping feeds. Use consistent GTINs so a model can resolve the same product across sources. Prune discontinued SKUs promptly, because an agent that recommends an out-of-stock item wastes the shopper's trust β and yours. Do this on a schedule, since a catalog drifts as products, prices, and inventory change.
Gotcha: watch for currency and unit mismatches across markets. A price in the wrong currency or a size in the wrong system (US vs EU) silently disqualifies you from the queries you'd otherwise win.
Step 5: Monitor which SKUs win the AI card
Optimization without measurement is guessing. The decisive question for a catalog is which SKUs actually appear in the AI product card, and whether you or a reseller wins the sale when they do. That's Share of Card: how often your listing appears in the AI product card versus a retailer reselling the same item, often at a different price.
Track card activation rate (whether a SKU gets a card at all), which products are listed, and where retailer quotes intercept your traffic. GEOly's catalog analysis and Share of Card surface exactly this across your whole catalog, and because GEOly carries industry-level shelf data, you can see category white space and how competitors' SKUs are positioned, not just your own. Start a free three-day trial at `app.geoly.ai` and run the 29-check GEO audit for a prioritized attribute fix list. The [Share of Card explainer](/blog/share-of-card-metric-shopify-brands-ai-shopping) shows how to read it.
Common mistakes
- Leaving feed fields empty because "the description covers it." Agents match on structured attributes, not prose. - Inconsistent data across page, feed, and marketplace β conflicting facts make the model hesitate. - Inventing ratings or return terms in schema to look competitive; models cross-check and penalize. - Optimizing attributes once and never re-auditing as prices, inventory, and SKUs change.
How to verify
- Run a sample of product URLs through a structured-data tester and confirm `Offer`, `AggregateRating`, `MerchantReturnPolicy`, and `FAQPage` parse without errors. - Audit attribute completion in your feed and confirm high-priority fields (material, size, fit, GTIN, use case) approach 95% filled. - Ask ChatGPT and Perplexity a constraint-heavy buying question and check whether your matching SKU appears at the correct price. - In GEOly, watch card activation rate and Share of Card by SKU to confirm the lift and catch retailer interception.
FAQ
How complete does my catalog data really need to be? Aim for 95%+ on the attributes shoppers actually filter by β material, size, fit, color, GTIN, use case. Missing fields don't just lower ranking; they exclude you from constraint-based queries entirely.
Is structured data enough, or does the on-page text matter too? Both. Schema gives the model reliable facts; clean, attribute-led text and an on-page FAQ give it quotable context. Agents use them together.
What's the difference between this and keyword SEO? Keyword SEO targets phrases people type. Catalog optimization for AI targets attributes and relationships an agent reasons over, so it can match complex natural-language constraints.
How do I optimize thousands of SKUs efficiently? Fix the systemic gaps first β a theme that omits `MerchantReturnPolicy`, a feed missing `material` β since one template change propagates across the catalog. GEOly's audit flags these patterns at scale.
How does GEOly fit in? It shows which SKUs win the AI card, which attributes are missing, and where resellers intercept you. Learn more via [what GEOly is](/blog/what-is-geoly-ai) and the [GEOly AI author page](/blog/author/geoly-ai).
Related reading: SKU-level AI visibility tracking β measuring which products AI recommends.
Related reading: AI shopping monitoring β per-SKU appearance history across engines.
Related reading: the KPI definitions and formulas behind Share of Card and Share of Model.

