Engines do not answer from memory. They retrieve pages, read them, and compose an answer out of what they found — and every one of those pages is recorded. Citino collects the URLs cited across all 1,785 answers in the category and ranks them by whether you can act on the page, influence it, or only watch it.
A visibility number tells you that Crateline is named three times as often as you are. This is the screen that tells you what it is being named out of.
Each row is a URL with its evidence attached: which engines cited it, on which of the 54 prompts, how often across the window, who owns it, and the date each engine first retrieved it.
The two columns are the same ledger under two sorts. On the left it is sorted by citation frequency, which is how this data is usually shipped. On the right it is sorted by movability, which is how it is shipped here.
Frequency puts the immovable pages on top. The most-cited page in almost any category is a large third-party comparison or a competitor's own landing page, so a frequency-sorted list opens with things you cannot touch and buries the directory listing you could have fixed before lunch. The engines' attention is not the scarce resource in your week; your attention is.
Same rows, same data, different order. Only one of them is a to-do list.
Yours. Pages on your own domain. You can rewrite one this week and know precisely when it went live. The 3PL billing page is one of these, and it is the page the live experiment is running on.
Third-party, influenceable. Directories, review sites, roundups, forum threads. You do not control them, but a listing can be corrected, a spec sheet updated, an outdated review answered. Slower than a rewrite, and it works often enough to be worth ranking.
Competitor-owned. Crateline's own comparison page. Cited constantly across the category, genuinely influential on what the engines say about you, and there is no version of your quarter in which you change a word of it.
Ownership is stored per row when the source is collected, and you can correct a misclassification. The competitor-owned panel is evidence. It is not work.
in the ledger · out of the denominator
The movable share is the fraction of the pages behind your losses that you could actually act on. It is a small number, and it is deliberately smaller than it could be.
Include competitor-owned pages and the figure inflates in exactly the wrong direction: every time Crateline publishes a strong comparison page and the engines start leaning on it, your movable share would rise. A metric that improves when your situation worsens is not a conservative metric, it is a broken one — and it would be the most quotable number on the screen.
So the rows stay, because they explain why you lose, and they stay out of the arithmetic. Citino reports a lower movable share than a tool that counts everything cited. That gap is the difference between a number you can plan against and a number that reads well in a screenshot.
On a page you own, the row carries Write brief →. On a third-party page you can influence, it carries Draft outreach →. On a competitor-owned page the action cell is empty. Not greyed out, not a lock icon, not a tooltip reading unavailable, not a prompt to upgrade. Nothing is rendered in that cell at all.
A disabled control is a promise with a delay attached. It tells the customer the thing exists and they have not yet found the route to it — and the honest answer is that it is never going to be true. We cannot edit Crateline's comparison page, and neither can you. The absence says it once, on the first row, permanently.
Every row records when each engine first cited that URL. It answers the question that follows every publish: did anything happen?
Visibility will not move for a while after you ship, and in that interval two different failures look identical on a chart. Either the engines have not picked the page up — a retrieval problem, and rewriting the copy a second time will not touch it — or they picked it up and it did not help, which is a content problem. A blank first-retrieval date two weeks after publishing tells you which one you have.
The live experiment on the 3PL billing page has its next read on 5 Oct; the first retrieval arrived before that, which is why the experiment is measuring rather than waiting.
Retrieval first, ranking later. → Act
A cited URL is a page the engine leaned on while composing that answer. It is not proof the page caused the recommendation, and the engines do not cite everything they read. Some answers cite nothing at all and are plainly built on something.
The ledger is therefore a record of where the engines showed their work. That is a great deal more than a visibility number and a great deal less than a causal model.
Copilot and Google AI Overviews are not in here. Neither exposes an official answer API, so any source list we published for them would be scraped, and a scraped row would sit in this table looking exactly like a measured one.
Pick a category and the ledger is populated on the first screen, already sorted by what you can change.