A shopper opens ChatGPT and types, "best merino wool base layer for winter running under $120." ChatGPT names four brands and links three stores. Your store sells exactly that product, and it is nowhere in the answer. This is the new shelf. This playbook covers what makes an assistant name one store over another, the concrete fixes that get you on the list, and before-and-after examples of AI-ready product data.
What makes AI recommend one store over another
Three page-level conditions decide whether an assistant can name you. Each is checkable:
- The product page exposes the matchable fields. Assistants match on
name,price,priceCurrency,availability, and category attributes (material, size, use case). A page missingOffer.priceorOffer.availabilitygives the assistant nothing to match against "merino, under $120," so it is skipped. Pages that carry these as structuredProduct+Offerschema are eligible; pages that carry only benefit adjectives ("premium," "ultra-comfortable") are not. - One page is the clean answer to the exact question. Assistants cite the page that answers "best X for Y under $Z" in liftable sentences. If no store page answers it, the assistant falls back to a third-party listicle — which is why review sites, not stores, win those answers today.
- The claims are named and checkable, and corroborated off-site. A store with thin data and zero third-party mentions is hard to recommend with confidence. Named specs plus a mention on a source the model already cites is what tips it.
You are not fighting for a keyword rank. You are supplying enough named, matchable detail for the model to pick you for a specific buyer question.
The fixes that work
1. Rewrite descriptions to name material, use case, fit, and price
Assistants match the words shoppers use. Name the use case, material, fit, constraint, and price band directly.
Before: "Our premium base layer delivers unmatched comfort and performance for every adventure."
After: "Merino wool base layer for winter running and hiking. 180 gsm, itch-free, regulates temperature to about -5°C. Men's S–XXL. Machine washable. List price $95."
The "After" version answers "merino base layer for winter running under $120" because it names material, weight, use case, temperature range, sizes, care, and price — seven matchable attributes in one block.
2. Add these exact schema types with real values
Give every product page:
- `Product` —
name,brand,sku,gtin(ormpn),material,size. - `Offer` —
price,priceCurrency,availability(InStock/OutOfStock),priceValidUntil. - `AggregateRating` —
ratingValue,reviewCount(only if real reviews exist). - `FAQPage` — on pages where shoppers ask concrete questions (sizing, compatibility, returns).
A page with Offer.price and Offer.availability populated is readable to an assistant; a page missing them forces the model to guess or skip you.
3. Publish one liftable page per high-intent question
For each high-intent question in your category — "X vs Y," "best X for Z," "what size X for a 6-foot runner" — publish one page that answers it in self-contained sentences.
Before: "Choosing the right base layer can feel overwhelming, but we're here to help."
After: "For winter running below freezing, choose 180–200 gsm merino. Below -10°C, layer a 250 gsm mid-weight over it. For high-output efforts, drop to 150 gsm to avoid overheating."
Every sentence in the "After" is a named object + a numeric threshold an assistant can lift and attribute to your page.
4. Earn a mention on the sources AI already cites
When AI answers "best X for Y," it pulls from third-party roundups and communities. Do this: search ChatGPT and Perplexity for your top 10 category questions, list the domains they cite, and for the 3–5 that publish "best X" roundups, send the editor your product's named specs (material, price, key attribute) for accurate inclusion. One correct line in a source the model already cites is a direct path into the answer.
5. Measure the exact questions you lose
Probe ChatGPT, Perplexity, and Google AI on 20–50 real buyer questions in your category. For each, record: do you appear, does a competitor appear instead, and what the answer says about you. That gap list is your fix backlog, ranked.
The action plan
- Measure — probe 20–50 category questions across ChatGPT, Perplexity, Google AI; log where you are absent and who is named instead.
- Fix the catalog — populate
Product+Offerschema on your top 20 products; add the missingprice,availability,gtin,material,sizefields. - Publish — ship one liftable page for each of the top 5 questions you lose.
- Corroborate — get named accurately on the third-party sources AI cites for your category.
- Re-measure — re-probe the same questions and connect any lift to visits and orders.
Where Arenza helps
Arenza runs this loop as one system. It probes ChatGPT, Google AI, and Perplexity on your category's buyer questions and reports an AI Visibility Score (are you found) and an AI Commerce Score (are you recommended and sold). Its AI Storefront view shows each product the way AI agents read it, the Products audit names the missing fields blocking a recommendation, Boost publishes GEO pages for the questions where you are absent, and Revenue attributes the AI-driven visits and orders back to your store. You can read your two scores on a free measurement tier before deciding what to fix first.
FAQ
How long does it take to get recommended by ChatGPT?
Catalog and schema fixes are read on the assistant's next crawl of your pages, often within days. Earning citations on third-party roundups and ranking newly published pages takes weeks. Re-probe your 20–50 question set after each fix so you can see which change moved which answer, rather than waiting blindly.
Is getting recommended by AI different from SEO?
Yes. SEO optimizes for a ranked list of links. AI recommendation optimizes for being named inside a written answer, which depends on readable Product + Offer schema, one clean page per buyer question, and corroboration from sources the model already trusts.
Do I need to publish blog content, or is fixing product pages enough?
Fixing product pages (schema plus matchable fields) makes you readable and eligible. Publishing one page per buyer question is what gets you named for those questions. Most stores need both.
Which product fields matter most?
Offer.price, Offer.availability, and the category attributes a shopper filters on (material, size, use case). A page missing price or availability is the most common reason an assistant skips a store it could otherwise recommend.