Shelf / Checks

19 checks

What an agent looks for, one check at a time.

Each check is something we can measure from a store’s own public data, and each one changes whether a product can be retrieved, matched, compared or trusted. Failures are worth reading in order.

Retrieval

4

Whether an assistant can reach and read your pages at all. Nothing below this matters if it cannot.

Blocking AI crawlers in robots.txtIf GPTBot, ClaudeBot or PerplexityBot are disallowed, your products cannot appear in the answers those systems give.No reachable sitemapThe sitemap is how a crawler discovers pages that are not linked from your navigation.No llms.txtAn emerging convention for pointing agents at the parts of your site worth reading. Low cost, currently low impact.The public product feedShopify publishes /products.json on every storefront unless it is deliberately blocked. It is how this audit runs without an install.

Structure

4

Whether the facts an assistant needs are machine-readable rather than implied by the layout.

Missing Product structured dataWithout Product or ProductGroup JSON-LD, price and stock have to be inferred from your layout rather than read.No GTIN, MPN or brandIdentifiers are how an agent works out that your listing and a competitor's are the same physical product.Ratings not exposed to machinesReviews rendered only as pixels cannot be read by an agent ranking options on quality.Price or availability missing from structured dataA Product block with no complete offer tells an agent the page exists but not whether it can be bought.

Substance

5

Whether anything on the page is specific enough to be worth quoting.

Duplicate product descriptionsCopy shared across many products reads as a supplier feed, and gives an agent no basis to prefer one of your items over another.The description describes a different productWhen the body copy never mentions the item in the title, the listing is either mis-imported or wrong, and both are treated as misrepresentation.Descriptions with no measurable specificationsAdjectives cannot be compared. Numbers can. Copy with no figures gives an agent nothing to filter or rank on.Keyword-stuffed product titlesA title built for 2015 search engines cannot be parsed into brand, product and variant by anything reading it today.Descriptions too short to evaluateUnder about forty words there is not enough text to establish what the product is, let alone why it is the right one.

Trust

6

The signals that decide whether an assistant names you or quietly picks a competitor.

Published products that cannot be boughtRecommending an out-of-stock item is the fastest way for an agent to learn not to recommend you.Supplier boilerplate left in descriptionsPhrases like 'Dear customer, please allow 1-3cm measurement error' identify the listing as an unedited dropship import.No real brand per productUsing the store name as the vendor on every product removes the field an agent uses to group and trust your catalogue.Products with no categoryAn uncategorised product is hard to place in a comparison set.Variants without SKUsA variant with no SKU is hard to reference unambiguously.Products with no imageAn imageless product is usually an incomplete one, and it is treated that way.

Run all 19 against your store.