Citation share is the share of measured AI answers in which a brand is named: the answers naming you, divided by the total prompt-engine runs, across a fixed set of buyer questions and a fixed set of engines. It is the main metric of AI visibility because it measures the thing that matters - presence inside the answer a customer actually sees - rather than proxies like rankings or traffic. Prominence, accuracy and tone are measured alongside it, as first-mention share, factual accuracy and sentiment. This guide sets out a protocol anyone can run.
Why not just measure rankings or traffic?
Because neither answers the question a business owner is really asking: when my customer asks the AI, am I the name it gives? Rankings measure a results page a growing share of customers never scroll. Traffic measures visits, but AI answers frequently satisfy the user without a click - a business can be recommended constantly and see little of it in analytics. The only direct measurement is to ask the engines what your customers ask, and record what comes back.
What makes the measurement trustworthy?
Discipline. AI answers vary between sessions, so a casual one-off prompt proves nothing in either direction. Three controls make the numbers meaningful. A fixed prompt set: the same questions, worded the same way, every period. A fixed engine set: the same engines every time - ChatGPT, Gemini, Claude and Perplexity are tested directly, with Google AI Overviews and Copilot visibility measured separately through Search Console and Bing Webmaster Tools. Fresh sessions: every prompt runs in a clean session with no history, because prior conversation contaminates answers.
How do you build the prompt set?
From real buyer language, not marketing language. Fifteen to twenty-five questions is enough for a small business, drawn from what customers actually ask: recommendation queries (“best implant dentist in Plymouth”), comparison queries (“X vs Y - which is better for nervous patients?”), and informational queries you should own (“how much do veneers cost in the UK?”). Crucially, split the set into branded prompts (your name appears in the question) and non-branded prompts (it does not). A business cited only on branded prompts is visible to people who already know it - and invisible in genuine discovery, which is where new customers come from.
How do you score an answer?
Record four measures per prompt, per engine: citation share - whether you are named at all; first-mention share - whether you are named first, which provides a consistent indicator of answer prominence; factual accuracy - whether what is said about you is correct (an AI recommending you with a wrong address is a visibility problem of its own); and sentiment - whether the treatment is positive, neutral or negative. Record who else is named too, because your competitors’ citation share is the context that makes yours meaningful.
non-branded share reported separately - it is the honest number.
Citation shares across competitors do not total 100% - a single answer can name more than one business, so treat each share as that business’s own presence rate, not a slice of one pie.
The citation-share protocol, end to end
- Build a fixed set of 15-25 real buyer questions, split branded / non-branded.
- Run every question on every engine in fresh sessions, on a fixed cadence - monthly suits most businesses.
- Score each answer: named? named first? factually accurate? sentiment? who else appears?
- Compute citation share: the percentage of prompt-engine runs in which you are named. Report non-branded share separately - it is the honest number.
- Compare against your baseline and your competitors, question by question - because visibility is won and lost per question, the per-question view tells you where to work next.
- Corroborate with AI-referral attribution in analytics: traffic and enquiries arriving from assistant surfaces confirm the visibility is producing commercial outcomes.
What does good movement look like?
Three patterns, in rough order of appearance. First, accuracy improves: the engines stop getting your facts wrong - usually the fastest win once entity and structured-data work lands. Second, informational citations arrive: you start being quoted for questions in your specialism. Third, and slowest, non-branded recommendations grow: the engines begin naming you when nobody asked about you by name. That third pattern is the commercial one, it is driven mostly by independent authority, and it compounds - which is why authority work starts early and why measurement never stops.
This protocol is exactly what an Answari engagement runs: the Visibility Audit establishes the baseline, and the monthly Watch reports movement against it. To see a single-question preview of your own numbers, start with the free visibility check.