Does each page answer what it is shown for?
What it is
On each project page, the card "Does each page answer what it is shown for?" grades the pages we published to your own site. Two or three AI models from different companies, with web search off, are each asked to quote the passage that answers the page's question. Every quote is checked against the page's own text, so a model cannot pass a page by agreeing with it.
Two kinds of result, never added together
OBSERVED-QUESTION ASSESSMENT — Search Console reported a real search that showed your page. That search is the question. It carries a verdict (Answers, Does not answer, or Contested when the models disagree), how many graders stood behind it, and the Search Console row: how often it was shown and at what position. TOPIC DIAGNOSTIC — there is no such search yet, so the page is asked about its own subject. It carries NO verdict, because a page is written to answer its own subject and passing that says little. It carries what is useful: the lines a writer could quote (checked against the page), the kind of source the models think the question expects, and what they say is missing. The headline counts pages analysed and splits them. There is no percentage, on purpose.
What the shape badge means
"Shape agreed by the models" on a diagnostic means their majority, reading the page afterwards, judged it the right kind of source for its topic. It is their opinion, it is labelled as one, and it is shown in grey on purpose: asked again about the same page they split more often than not, so it is never a pass. A diagnostic has no verdict to pass. "Not the right shape yet" is the same opinion the other way, and worth reading because it names what a rewrite would add. Whether asking the shape questions in the same call moves a grader's verdict was tested on 2 September 2026: 37 documents, three graders, each document graded with the questions, without them, and with them again as the noise floor. No verdict flipped (0 of 30, 0 of 31, 0 of 35). The test is repeated whenever the prompt changes.
Why a page has no search yet
The card says why, because the reasons need different fixes. "Google reported no search for this page in the last 90 days" is about the page. "Search Console is not connected", "no property covers this domain" and "we could not read Search Console" are about the connection, and only connecting it changes them. Search Console reports up to three days ago, so a page published this week always reads as "no search yet".
Shape-first: asking before writing
On every project, the order is reversed. Before anything is written for a topic, two or three models from different companies are asked what kind of content would actually get cited for it, which concrete things it must contain, and where else content in that shape belongs. Their majority answer is stored as a contract for the topic. The page is written to that shape, carrying those elements, and then checked: each element must be quoted from the page, and the quote must be found in it. If any are missing the page is redrafted once, naming them, and published either way. Every topic competes on your own site unless you switch it. Where the panel thinks the topic is usually answered by somebody else's kind of page — a review, a roundup, a comparison — the page is still written for your site in the shape that wins, a comparison or list where that is what gets cited, honest that it is your page. The places the panel named are shown on the card as "Also worth being on", as advice. The panel is told which of the pages it sees are your competitors' own (sites you labelled Competitor under Sites AI cites, or on your watchlist's competitor list): a topic answered mostly by competitors' own roundups and tool pages is one your site can take part in too. Where AI answers cite free tools and product pages for a topic, the article is published as normal, with the suggestion to make your own tool or product page answer the question as advice. "Place it elsewhere", on the topic in your topic list or on the card, is the other choice. Then the panel's placement becomes one task only you can do, beside the Reddit threads, naming every venue it suggested: they are alternatives, and one placement is the whole task. The page on your site is written as the nearest shape it can be, and a page for a topic answered by free tools is held for you with that reason. "Compete on my site" switches it back. Every placement task also has "Compete on my site instead": the topic competes on your site, the task is set aside, and you are told where the topic's page on your site is, or offered to build it now. Undo puts it back. Reopening a placement task you set aside sets its topic to "Place it elsewhere". On the card such a page leads with the contract: "Written as a Comparison, carries 4 of 5", with each element ticked or crossed. The graders' later opinion of its shape is shown underneath, as an opinion. The contract is checked; the opinion is not.
Rebuilding a page in the right shape
On a shape-first page the button reads "Add what the panel asked for" and hands the writer the contract: the decided shape and the elements the check found missing. Elsewhere, where the models say a page is not the right shape, "Rebuild in the right shape" hands their brief to your default AI writer with the page's own text as the only source. The rewrite keeps the verified lines word for word, may add no facts, and may link only sources already on the page or recorded against your business in the Entity Factory — never an invented reference. On a connected WordPress site the post is updated in place at the same address. Elsewhere the rebuilt page is held on the card for you to copy across. Either way it is graded again on the next nightly sweep, so before and after are two readings. A page an AI engine already cites is the exception: its text is never replaced until you approve the change. The row says so before you press the button, and the rewrite then waits under "A page update waits for you" at the top of the project page, as a Re-align does, for you to read and publish in place or discard. If you leave it, the page stays as it is. Send again on a rebuild that never reached its page is not an approval either: on a page that is now cited, it moves the rebuild there instead. A rebuilt page is one text, so an update from new material that was waiting for the same page no longer waits once the rebuild does. A re-align of the same page can wait beside it - the row says so - and each replaces the whole page, so publish the one you want. Once a page is settled - you chose to leave it as it is, or it was rewritten (here, by Re-align or by a page update) and graded again - it stays in the list for two days so you can see the outcome, then moves under "Done" at the foot of the list, with what was decided and when. The counts at the top still cover every page.
What it costs
Grading is off unless you switch it on (Settings → Answerability grading), because it runs on your own AI keys: two or three calls per page, once, nightly, newest first, up to ten pages a night. A page is graded again only when the question changes, our text extractor changes, or you rebuild it. The nightly check that a verified passage is still on the page reads the page once, uses no AI, and runs whether grading is on or not. A rebuild is one call to your default writer. The shape of every page is decided with your providers before it is written, whatever this switch says — that is the check that comes first; this is the optional one afterwards. Two other readings spend your key the same way and are also off unless you turn them on: before-readings for new gaps and the passage check after writing. Their calls are listed under What measuring costs.
Answerability says nothing about whether an engine will surface the page. Visibility is measured separately, per engine, and the two are never averaged.