Product availability, Merchant listings, crawler access, and review policy — a checklist for stores that want to be named.
A shopper asks an assistant for "waterproof trail runners under $120 with wide sizes." Your product exists. Your HTML hides the size chart behind a widget, the bot gets a 403 from the WAF, and the answer cites a retailer with uglier photos but clearer specs. Ecommerce AEO is mostly access, product facts, and merchant honesty — not a new keyword tool.
Use this checklist for stores that want to be named. Work top to bottom; do not start with blog content if crawlers cannot see PDP HTML.
Priority checklist
| Priority | Area | Pass criteria |
|---|---|---|
| P0 | Crawler access | Answer bots allowed in robots; WAF not 403ing them on PDPs/category |
| P0 | Indexable HTML | Title, price, availability, brand in raw HTML |
| P1 | Product identity | Clear name, brand, SKU/GTIN where applicable |
| P1 | Merchant honesty | Stock and price match what checkout charges |
| P2 | Reviews policy | Real reviews; no fabricated AggregateRating |
| P2 | Policies | Shipping, returns, contact visible |
| P3 | Category content | Useful guides without thin doorway spam |
P0 — Access and rendering
- Confirm
GPTBot,ChatGPT-User,PerplexityBot, and other answer crawlers you care about are not Disallow’d on product and category paths. - Allowlist those agents in Cloudflare / WAF if Bot Fight Mode is on — same class of fix as allowlisting AI user agents.
- Verify PDPs return 200 with product name and price in the first HTML response (SSR, static, or prerender). If the first byte is an empty shell, fix rendering before writing more copy — see SSR vs CSR for AEO.
- Eliminate soft 404s on retired SKUs; redirect or return a real 404 — soft 404s confuse AI crawlers worse than clean misses.
Know which bot you are welcoming. ChatGPT-User vs GPTBot is a common mix-up: blocking the live-answer fetcher while allowing the trainer (or the reverse) wastes the whole exercise.
P1 — Product and merchant signals
| Field | Why assistants need it | Tip |
|---|---|---|
| Product name | Citation text | Match storefront and schema |
| Brand | Entity resolution | Same string sitewide |
| Price + currency | Decision answers | Visible, not image-only |
| Availability | Avoid recommending empty stock | Sync feed and page |
| Size / variant facts | Long-tail Q&A | Tables beat carousels for bots |
| GTIN / MPN | Identity across merchants | When you have them |
Use Product schema with real offers. Do not invent ratings. Merchant listing feeds (where you use them) should not contradict the PDP — conflicts teach models that you are unreliable. For field choices on software-like goods, the spirit of Product schema for SaaS still applies: name, description, brand, offers — no fake stars.
Hypothetical: say you run 20 category prompts about "wide size trail runners." If your wide sizes only appear after selecting a variant in JS, the assistant may never see them.
P2 — Reviews and trust
Claim profiles on the review sites your category actually uses. Respond to reviews. Do not spray fake 5-star schema. Assistants already cite major review destinations; fighting them is wasted motion compared to review sites AI keeps citing.
On-site Q&A is fine when questions are real. Mass-generated FAQ spam on every PDP is noise.
Return and shipping policies belong in crawlable HTML. Assistants answering "Can I return these in 30 days?" should not have to invent your policy from a competitor.
P3 — Content that supports shopping answers
- Size guides, materials, and care pages with stable URLs.
- Comparison tables for "X vs Y" within your catalog — honest, not sludge (comparison pages).
- Avoid doorway "best [city] [product]" spam.
- Keep brand and product entities distinct if you sell multiple lines (multi-brand entity separation).
Step-by-step: one-day store pass
- Curl or text-fetch 5 PDPs and 2 categories as a bot would.
- Fix any 403/JS-empty responses before editing copy.
- Align Product schema with visible price and availability on those URLs.
- Add or repair brand/Organization identity on the homepage.
- Update robots/WAF allowlists; retest.
- Submit honest sitemap
lastmodfor changed PDPs. - Spot-check two shopping prompts in an assistant; note factual errors; fix sources.
- Re-scan with BrandKnown to confirm access and clarity checks flipped.
Metrics without self-fooling
Similarweb Gen AI Landscape 2025 (US desktop, Sep 2025) reported higher average engagement (~15 min and ~12 pages/session for ChatGPT referrals vs ~8 min and ~9 for Google) and about 7% conversion on transactional sites from ChatGPT referrals versus about 5% from Google in their cut — interesting, not a guarantee for your store. Track AI referrals in GA4 with sober sample sizes (track AI referrals in GA4). SparkToro / Similarweb (Jan–Apr 2026) put Google zero-click near 68%; many shopping journeys never hit your analytics.
Honest ceiling
Marketplaces and incumbents dominate many product answers. Your controls are access, accurate offers, and clear brand/product entities. A BrandKnown scan (~60 seconds, free first scan) flags readiness issues on the site you control. Agency ($49/mo) helps if you manage multiple storefront brands. Re-scan after fixes to prove the checklist moved — not to promise you will win the next shoe query.