The Citation Gap, in one number

Here is the whole idea.

For every URL, hold two numbers next to each other: how often AI agents fetched it, and how many people arrived at it from an answer engine. Divide the first by the second.

ratio = agent_fetches / answer_arrivals

That ratio is the Citation Gap. Two thousand fetches and four arrivals is a page being taken and not returned. Two hundred fetches and ninety arrivals is an exchange that is working.

No other analytics tool reports it, and the reason is not that nobody thought of it. It is that the two halves live in different places. The taken half is in your web server’s access log, which analytics tools do not read, because crawlers do not run JavaScript. The returned half is in the referrer of an ordinary pageview, which analytics tools do read but file under “referral”, next to a link from someone’s newsletter.

Put both in the same store, keyed by URL, and the ratio falls out.

Thresholds, published

A verdict that cannot be checked is an opinion. So:

• cited means at least three answer-engine arrivals in the window
• heavily crawled means at least twenty agent fetches in the window
• a cited, heavily crawled page is extracted when the ratio is above twenty-five

A page with two arrivals and nineteen fetches gets no verdict at all. Thin data earns a shrug, not a diagnosis.

What it cannot see

This is the part a product would normally leave out, which is precisely why it is here.

Referrers usually arrive as an origin. Browsers default to strict-origin-when-cross-origin, so a cross-origin referrer carries a host and nothing else. Any answer surface sharing a host with an ordinary search page cannot be told apart from that site’s own results in the common case. Counting every DuckDuckGo search as an answer-engine arrival would inflate this product’s own signature metric with a number nobody measured.

In-page answers are invisible. A reader who clicks a citation inside an AI overview arrives with the search engine’s host as the referrer, indistinguishable from an ordinary search click.

So the returned half is a floor, never a ceiling. The real exchange is at least as good as the number shown, and the interface says so rather than implying a precision it does not have.

What to do with it

Nothing, necessarily. extracted is a measurement, not an instruction.

Blocking the crawler is one response. Writing the page so an answer has to link out is another. Deciding that the reach is worth the trade is a third, and it is often the right one: plenty of writing exists to be read, not to be clicked.

What is not reasonable is not knowing. Until the two numbers sit next to each other, the question cannot even be asked, and it is being answered for you every day by a system that does not consult you.