How citelity measures AI visibility

Where each number comes from

Every engine is included on every plan. There are no per-engine add-on fees, so nothing about the list below depends on what you pay.

What the badges mean

Every metric carries a badge describing how it was obtained. The badge is part of the number, not decoration around it.

How a citation is counted

An engine typically looks at more pages than it ends up citing. We count only the sources the answer actually cites. Counting everything the engine merely consulted would inflate your citation rate and make the number useless for deciding what to fix next.

Citation position is the order your page appears among the cited sources for that answer. Where an engine attaches sources to individual passages, position follows the order of the answer itself, because the sources behind the opening lines are the ones a reader encounters first.

Mentioned without a link is tracked separately. Being named in an answer without a citation is a different situation from being cited, and it calls for a different fix, so the two are never merged into one figure.

When an engine returns no AI answer for a prompt, that is recorded as a result, not an error. Plenty of queries never trigger one.

How often we check

A full scan runs weekly across every tracked prompt and every engine. On Pro and Scale you can star individual prompts for a daily pulse on ChatGPT and Google AI Overviews.

Daily checks on everything would cost more and tell you less. LLM answers fluctuate from one day to the next without anything having changed on your site or theirs. Weekly depth plus a daily pulse on what matters gives you the trend without turning normal noise into a stream of alarms.

How we measure whether a fix worked

After a page is created or updated, we take a snapshot of its metrics at that moment and measure against it three times.

The staging is not caution for its own sake. Content changes usually begin moving positions at four to six weeks and settle between eight and twelve. Judged at two weeks, most legitimate work would look like failure — measured by a clock rather than by the work.

If day 60 shows no improvement, that is what your timeline will say. You would see the same thing in your own Search Console, and a tool that hides it is only a slower way of finding out.

What we don't claim

We do not publish an accuracy percentage, and we are wary of anyone in this category who does. Doing so honestly would require a ground truth to compare against — a complete record of every AI answer served to every user. No such record exists outside the engines themselves, and they do not publish it.

So we do not tell you how close our figures are to a total we cannot see. We tell you precisely how each figure was obtained and let you judge it. That is the honest ceiling in this category today, and claiming more would mean inventing the missing part.

We also make no promises about outcomes or timelines. What is under our control is the process: finding the prompts and pages where AI answers without you, writing the fix, and measuring what changed.

Questions