AI search is new enough that most teams either measure nothing or measure everything — twelve dashboards, no decisions. This is the KPI stack we've settled on after tracking brands across categories: five numbers, each answering one management question, in the order they should mature.
The KPI stack, by what each layer answers
Five numbers, one story — ordered from headline to leading indicator to money
Citation share is the layer that predicts the others — it registers your off-site work weeks before the answers do.
What % of AI answers to your tracked buying questions name your brand?This is the category's share-of-shelf. Measure it across a stable prompt set (how to build one), across multiple engines, on a cadence — because single runs are noisy1 and the engines disagree with each other.2 Report it as a trend, never a snapshot. This is the number that belongs next to your rankings in the board deck.
Of all brand mentions in your category's answers, what share is yours? Visibility rate can rise while you lose ground — if a competitor rises faster. SoV frames every result competitively, which is how buyers experience it: they don't see your mention count, they see a shortlist with you or without you. Watch the ratio between you and the category leader; closing it is the actual game.
Being named first in a list of five is a different commercial event than being named fifth with a caveat. Track average position when named, and how you're described — positive, neutral, hedged. The sentence AI uses about you is your new first impression, repeated at scale. This layer usually explains the “we're mentioned but not chosen” mystery.
Of the sources AI cites in your category, how many mention you? Mentions follow evidence — web-wide brand mentions correlate 0.664 with AI visibility3 — so your presence in the cited pages moves beforeyour visibility rate does. When a listicle adds you or a review push lands, citation share registers it weeks before the answers reliably do. It's the KPI that makes the others predictable, and the one that assigns work: every citing page you're absent from is a task with a URL.
AI-referred traffic is small for most brands today but converts at a serious premium — Semrush measured AI-search visitors as several times more valuable than organic ones,4because they arrive pre-sold. Tag and track: sessions from AI referrers, “how did you hear about us” answers that say ChatGPT, and pipeline sourced from AI-visible questions. Early on the sample is thin — treat it as directional evidence for the program, not a channel report.
Two traps. One-off audit scores— a single “AI visibility score” captured once is a photo of weather; the volatility data says it may be different tomorrow.1 Vanity breadth — tracking 500 prompts nobody reads beats 25 prompts everyone acts on exactly never. Keep the stack small enough that when a number moves, someone asks why — that question, asked weekly, is the entire management system.
Maturity path: start with visibility rate and SoV (week one), add citation share when you start doing off-site work (it attributes your effort), add position/sentiment as mentions appear, and stand up outcome tracking once AI referrals are a visible trickle. Five numbers, one story: are we becoming the answer?
If you'd rather not assemble this stack from spreadsheets: Beacon ships it out of the box — visibility rate, share of voice, position, sentiment, and citation share, tracked daily across ChatGPT, Gemini, Perplexity, Copilot, and Google AI Mode, with every movement traceable to the sources that caused it.
Beacon tracks visibility, share of voice, position, sentiment, and citation sources daily across five AI engines — one dashboard, no assembly.