A citation is a placement, not a ranking
When Google's AI Overviews or AI Mode answers a question, it links a handful of sources behind the generated text. Unlike a classic search result, there is no stable position ten to defend: the answer itself absorbs the click, and the citations behind it vary with the query, the locale, and the day.
That volatility is why single observations are dangerous. A brand can appear in an AI Overview on Monday and be absent on Wednesday with nothing changed on its website. The honest unit of measurement is the repeated, condition-recorded sample — same prompt, same provider, same region — not a one-off screenshot.
What a citation is worth
A citation inside an AI answer plays roughly the role a top-of-funnel mention plays in the old web: it makes the brand part of the consideration set at the exact moment a buyer is forming a shortlist. The value is real but indirect — most AI answers send fewer clicks than a blue link did, and some send none.
The practical way to value a citation is commercial, not volumetric: which question was answered, whether it describes a purchase decision, and whether competitors were named alongside you. A citation on "best CRM for small teams" is worth more than a thousand impressions on trivia.
How to measure it without fooling yourself
Record the prompt, provider, surface, locale, region, repetition, and timestamp for every sample. Repeat each prompt several times before believing a pattern. Keep the raw answer and the native citation URLs — not a summary a second model wrote about them.
Then recheck under identical conditions after you ship a change, and report what actually happened. That is the entire methodology behind Prefometry's Citation Score and Evidence Confidence: two numbers, one for value and one for repeatability, never blended.