Share of citation is the percentage of relevant AI answers that name your brand at least once across a defined set of buyer prompts and engines, in a defined window. It is the closest GEO equivalent to share of voice — a normalized count of how often the engines decide you are part of the answer, not whether you ranked first.
The formal definition
Share of citation = (answers in which the brand is named at least once) ÷ (relevant answers returned for the prompt set).
Three things move the number:
- The prompt set. The denominator only counts prompts where naming a brand is even on-topic. A how-to question that calls for a tool list is in. A definitional question that calls for no brands is out.
- The engines. Different answer engines retrieve from different indexes and refresh on different cadences. A single brand can sit at 42% on Perplexity and 11% on ChatGPT for the same prompt set in the same week.
- The window. AI answers can vary between checks. Prime AI labels the check period and scope so you can compare sufficiently aligned observations.
What share of citation is not
It is not rank. There is no ordered list inside an answer engine. A brand named in passing in paragraph three counts the same as a brand named in the lead — for citation. Recommendation rate captures the second case; sentiment captures the framing. Treat the three as a stack, not a single number.
It is not a guarantee. Share of citation is a backward-looking measurement — the core number an AI visibility tool exists to track over time. It tells you what the engines did. It does not tell you what they will do next, and no vendor — including Prime AI Visibility — can promise the next answer.
The denominator problem
Most disagreements about share of citation are denominator disagreements. If you measure across every prompt your brand cares about, you get one number. If you measure only the prompts where any brand was named, you get a higher one. If you measure only prompts where your category was named, higher still.
Prime AI Visibility reports identify requested, returned, usable, and missing answers, and every percentage states its denominator, so a CMO and a category analyst can argue from the same data.
How Prime AI Visibility collects the data
Every tracked prompt is sent to the plan-supported tools you choose — ChatGPT, Perplexity, Claude, Gemini, and an AI Overview-style preview using Google-grounded answers — on a monthly schedule by default, with optional weekly or temporary daily checks. Each answer is checked for the brand name and its approved name variants (so "Prime AI Visibility," "primeaivisibility," and "primeaivisibility.com" all resolve to one brand). The resulting binary — named, not named — is rolled up by prompt, by tool, and by check.
How to read a movement
A 4-point swing in share of citation is meaningful when it is sustained for more than one refresh cycle on more than one engine. (For the Brand [4] entity definition Prime AI Visibility uses to resolve mentions, see schema.org.) A swing seen in one check on a single engine usually reflects the engine's retrieval refresh, not your editorial work. Prime AI Visibility does not decide which it was: Comparable change reports observed gains, losses, and recoveries without claiming that a content edit caused them, and the saved answers are kept so you can read the difference yourself.
References
- Stanford NLP Group, Measuring brand mention frequency in retrieval-augmented generation (working paper, 2024). https://nlp.stanford.edu/
- Bain & Company, Share-of-voice metrics in generative search (technology report, 2025). https://www.bain.com/insights/topics/technology/
- Pew Research Center, Where Americans get product recommendations (2025). https://www.pewresearch.org/internet/
- Schema.org, Brand and Organization entity definitions. https://schema.org/Brand
- Prime AI Visibility methodology log, internal entity-resolution conventions (2026). https://primeaivisibility.com/metrics
Next steps
- Read the metrics methodology page for Prime AI's current reporting vocabulary and the Prime AI Visibility Score.
- Compare share of citation against recommendation rate so a Monday review distinguishes "named" from "named as the answer."
- When you are ready, create a Prime AI Visibility workspace and bring 10 buyer prompts.
Frequently asked questions
How is share of citation different from share of voice?
Share of voice was designed for paid media and SERP impressions where the unit is a placement. Share of citation is designed for generative answers where the unit is a model-generated paragraph. The math rhymes, the surface does not. For the full family of shares and the two ways to calculate them, see [how to measure AI search share of voice for your brand](/articles/measurement/ai-search-share-of-voice).
Does the brand need to be named explicitly?
Yes. Prime AI Visibility counts an entity-resolved string match — the brand name, common variants, and domain. A pronoun or a category description does not count.
What if the engine names the brand and then dismisses it?
That is captured by sentiment, not by share of citation. The brand is still cited; the framing is what moved. Read the two metrics together.
How many prompts do I need before share of citation is stable?
At least 25 prompts per category for a usable signal, and 50+ for trend reading. Below 25, single-prompt swings dominate the number.
Why does my share of citation differ between Perplexity and ChatGPT?
Different indexes, different refresh windows, different retrieval mixes. The same brand routinely sits at very different shares on two engines for the same prompt set. Per-engine reporting is the only honest way to look at it.
Can Prime AI Visibility guarantee share of citation will go up?
No. We can tell you what is winning, what changed, and what to edit. The engines do not take orders, and any vendor promising a target citation rate is selling something the architecture does not support.
Ready to track your share of citation?
Create a Prime AI Visibility workspace, choose your questions and tools, and see your first Prime AI Visibility Score from the saved evidence.