What can an answer engine read?

A great many sites clear the access bar and still never get mentioned. The crawlers arrive, read the page, and find nothing they can safely lift: no stated business type, no address a machine can parse, no services expressed as anything other than a paragraph of adjectives, no prices, no hours. The page reads fine to a person and offers an assistant nothing to quote.

This is the gap that structured data exists to close. Structured data is a block of machine-readable facts embedded in the page — usually JSON-LD — that states plainly what the business is, where it is, what it sells, and what people have said about it. It is not a ranking trick. It is the difference between an assistant inferring your business from prose and an assistant reading it as fact.

The tool fetches your public pages and evaluates four groups of signals. They are ordered by how much difference they make when they are missing.

• Structured data — is there JSON-LD on the page, does it validate, and does it describe the actual business rather than a generic WebSite object? Nine of the scanner's forty-three rules sit in this category, the largest AI-relevant group.
• Metadata — titles, descriptions, canonical URLs, and indexability. These decide how the page is summarised and whether duplicate versions of it compete with each other.
• Crawler access — whether the named AI crawlers can read the page at all. A perfect structured-data implementation behind a blocked crawler is worth nothing.
• Sitemap coverage — whether the pages you care about are discoverable in the first place, or exist only behind a link a crawler never followed.

When structured data is the problem, it is almost always one of three shapes, and the fix is different for each.

• None at all. The page has no JSON-LD. Common on hand-built sites and on templates where the theme author left it to a plugin that was never installed. The fix is additive and low-risk.
• Present but broken. There is a JSON-LD block, but it fails validation — a malformed date, a required property missing, a type that does not exist. Parsers discard the whole block. From the assistant's side this is identical to having none, which is why it goes unnoticed for years.
• Present, valid, and empty of meaning. The page declares a WebSite or Organization with a name and a URL and nothing else. It validates cleanly and says nothing an assistant could use to answer a real question. This is the most common shape on sites that installed an SEO plugin and considered the job done.

The third case is the one worth dwelling on, because the site owner usually believes the work is finished. Validation is a floor, not a finish line. The test that matters is whether someone reading only the structured data could tell what you sell, to whom, and where.

Work in this order, because each step is wasted if the one above it is broken.

• Confirm the search-side AI crawlers can reach the page. Nothing downstream matters otherwise.
• Fix broken JSON-LD before adding new JSON-LD. A block that fails to parse is discarded entirely, so adding more to a broken block changes nothing.
• Make the primary entity specific. LocalBusiness or a narrower subtype with a real address, real hours, and real services beats a bare Organization every time.
• Then extend to the pages that answer questions — services, pricing, FAQ — rather than adding more markup to the homepage.

Almost none of this requires a redesign. Structured data is added to the page template, and on most CMS platforms it is a settings screen rather than a code change.

This tool measures readiness, not outcome. It reports whether your site exposes the signals that support citation. It cannot tell you whether ChatGPT mentioned you this morning, how often you appear for a given question, or which competitor is being cited instead — those are monitoring questions, not page-inspection questions.

It is also a slice. Citation readiness draws on structured data, metadata, crawler access, and sitemap signals; the full scan additionally covers HTML markup, privacy and consent, security posture, email deliverability, and tech stack, for forty-three rules in total. The methodology page lists every one of them.