The AI-visibility page checklist

Answer engines lean on schema.org JSON-LD to identify the entity, its facts and its relationships. A page with none makes them guess.

Fix: add a <script type="application/ld+json"> graph describing the page entity — Organization, Product, Article or Service — plus the WebPage itself.

A block that fails to parse is ignored entirely. Invalid structured data is worth exactly as much as none, while looking like work was done.

Fix: validate every block and repair the syntax error. A trailing comma is the usual culprit.

Without a recognised type, nothing tells an engine what kind of thing this page is about.

Fix: declare one of Organization, Product, Article, Service, WebSite, WebPage, SoftwareApplication or LocalBusiness, with a clear name.

Question-shaped sections are the most liftable content on a page. Unmarked, an engine has to infer that they are questions at all. Pages with no Q&A content pass — there is nothing to mark up.

Fix: wrap the visible questions and answers in FAQPage JSON-LD that mirrors the visible text exactly.

Engines favour fresh sources. With no dateModified there is nothing to signal recency with.

Fix: add dateModified to the page's JSON-LD and keep it honest — bumping it without a substantive change is a low-trust signal and it ruins your own before-and-after readings.

Citation priority decays once a page goes stale. This passes when dateModified falls within roughly the last six months and is not in the future.

Fix: refresh substantive pages on a real cycle and let the date follow the work.

No H1 leaves the page without a stated main topic; several blur it for a parser.

Fix: use one H1 stating the page's subject, and demote the rest to H2 or H3.

A page that does not open with a direct "X is a …" definition offers no clean snippet to lift, so an engine writes its own — from whatever else it found.

Fix: start with a one-sentence definition: "The subject is a category that does what."

Models weight the first sentences most heavily. Naming the subject there, inside a focused block, makes the page far likelier to be quoted verbatim.

Fix: open with a block of roughly 50 to 120 words that names the subject in the first sentence and answers the core question directly.

Question-led sections map directly onto how people prompt an assistant, which makes them disproportionately citable.

Fix: add a short FAQ answering the questions buyers actually ask, each answer self-contained.

Dense prose with no list or table is harder to extract facts from — for a reader and for a parser.

Fix: put specs and feature facts in a real table or a bulleted list.

Under about 150 words in the raw HTML, there may be nothing to retrieve. If the content is rendered client-side, most engines will not see it at all.

Fix: make sure the substantive content is server-rendered rather than injected by JavaScript.

The datasheet pattern: the page links to PDFs, carries no table and almost no numbers of its own. Most engines retrieve PDFs poorly or not at all, so those facts are effectively invisible.

Fix: mirror each PDF's key facts as real HTML — a table on the page or a linked HTML spec page. Keep the PDF as the formal document, not the only copy.

Without Organization.sameAs, an engine cannot reliably tell your brand apart from a similarly named one.

Fix: add Organization JSON-LD with sameAs pointing at Wikidata, Wikipedia, LinkedIn and your official profiles.

An "also known as" line carrying alternate names, abbreviations and model numbers helps an engine match the way people actually refer to you — including the name you used before the rebrand.

Fix: add one sentence listing the alternates, and mirror them as alternateName in the structured data.

Unverifiable superlatives get discounted, and they crowd out the specific claims that would have been quoted instead.

Fix: replace each superlative with a specific, number-backed claim.

The title is the first thing that names the page's subject to a machine.

Fix: add a descriptive <title> naming the subject, answer-first.

Passes between 50 and 170 characters; aim for 120 to 160. Too short says nothing, too long gets cut.

Fix: summarise the page answer-first in about 155 characters.

Without one, duplicate URLs split the signals that should have accumulated on a single page.

Fix: add <link rel="canonical"> with the page's preferred URL.

Preview metadata decides what a shared link looks like — which decides whether it gets clicked at all.

Fix: add og:title and og:description.

Missing alt text hides an image's meaning from engines and from assistive technology alike.

Fix: add descriptive alt text to every meaningful image, and restate any in-image number in nearby prose.

A robots meta tag telling engines not to index the page ends the conversation — it cannot be cited however well optimised it is.

Fix: remove the noindex directive if you want this page cited.

A non-2xx status means engines may never fetch or index the page. Reported as not assessed when the HTML was pasted rather than fetched.

Fix: make sure the page returns 200 — repair the redirect chain or the error.

If robots.txt blocks the path, engines cannot fetch it. Worth separating the two decisions: blocking answer agents removes you from AI answers, while blocking bulk training crawlers is a licensing choice.

Fix: allow the path for the crawlers you want reaching the page. Rules to copy are in the templates pack, and the access checker tests each agent live.

A canonical pointing elsewhere hands the credit to that page instead. Sometimes deliberate; usually a template left on the wrong setting.

Fix: point the canonical at this page's own URL unless you genuinely want another page to win.

An emerging convention for pointing answer engines at your most citable pages. Cheap insurance, not a lever — it is a minor check here for exactly that reason.

Fix: publish one listing your important pages. Our generator builds it from your sitemap.