AI visibility metrics

The metric family

Five measurements cover most of what teams report, and each answers a different question.

• Citation rate — of the prompts in a defined set, in what share of answers does the entity appear? The base rate of visibility; per product, it is the foundation of the citation score.
• Share of model — of the answers in a category, what share features you versus competitors? The relative read; see share of model and its synonym AI share of voice.
• Answer position — when cited, are you the first recommendation or the seventh? Composed answers concentrate attention at the top even more than results pages did.
• Sentiment — how does the engine frame you when it mentions you? A citation that says "popular but frequently reported to pill" is not a win; see AI sentiment analysis.
• Readiness scores — can engines and agents retrieve and parse the page at all: schema completeness, crawler admittance, machine-readable price and availability. The agent-readability score is this measurement per product page.

Leading vs lagging, and why both are needed

Citation rate, share of model, position, and sentiment are lagging metrics — they report outcomes the engines already produced, and they move on the engines' schedule, not yours. Readiness scores are leading — they measure the controllable inputs that make future citations possible. A dashboard with only lagging metrics tells you that you lost without saying what to fix; only leading metrics, and you are grading homework nobody marked. The pairing is the point: fix what readiness flags, then watch whether citations follow over weeks, on a held-constant prompt set so the comparison is honest. The instrumentation underneath is prompt tracking, and normal week-to-week drift means trends matter and single readings do not.

The grain problem: brand-level numbers hide the revenue

Every metric in the family can be computed at brand level or at product level, and the choice decides whether the number supports action. A brand-level citation rate of 40% sounds healthy and can coexist with the three products that drive half of revenue never appearing in an answer — a brand mention doesn't tell you which SKU won. Product-level metrics resolve the question a merchandiser can act on: which products, which engines, which intents, which PDP fields. The argument in full: measuring AI visibility — what actually matters.

How eCommerce Insights instruments the family

Per SKU, per engine, on held-constant buyer-prompt sets across ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude, and Copilot. Each product carries two composite scores — the citation score (lagging: is it recommended) and the agent-readability score (leading: can an agent parse it) — with the family's raw reads (citations, position, sentiment, share against named competitors) underneath, and history so multi-week trends separate from drift. Methodology for the composites is documented in the PDP score reference. Engines change behavior on their own schedules — vendor documentation such as OpenAI's bot documentation defines what is controllable, and the metrics exist to measure exactly that controllable surface.

Related terms

• Citation score — the per-product composite of the lagging metrics.
• Agent-readability score — the per-product leading indicator.
• Share of model — the competitor-relative member of the family.
• AI visibility score — the 0–100 rollup across the family.
• Prompt tracking — the instrumentation every metric depends on.
• AI sentiment analysis — the framing read behind the counts.