What is citation share?

It answers a different question from every other AI visibility metric: are you the evidence the answer was built from, or just a name inside it? Those are separate things, they move separately, and conflating them is the most common reporting error in this field.

The formula

Both halves need pinning down before the number means anything:

• The question set must be fixed. Change the questions and you change the denominator, so the series is no longer comparable with itself. A report that does not print its prompt list cannot be trusted to have kept it stable.
• “Your own domain” needs a rule. Decide up front whether subdomains, a documentation site on a different domain, and your profile pages on third-party platforms count as yours. Our own analysis counts only domains we own; a G2 profile about us is not a citation of us.
• Decide whether to deduplicate. If one answer cites three different pages of yours, is that one citation or three? We count each cited source, so it is three. Either rule is defensible; only changing it mid-series is not.

A worked example, with real numbers

On 19 September 2026 we ran our own audit: eight buyer questions, put to Perplexity through its API, both runs stored in full. The figures below are from run A, the stored record 2026-09-19T04-22-26-292Z.json.

• Questions asked8
• Answers analysed8
• Total sources cited90
• Distinct domains cited (run A)74
• Distinct domains cited (run B)75
• Sources pointing at bungad.com0
• Citation share0.0%

Zero divided by ninety. There is no rounding to argue about, and that is the number we publish, because a company selling AI visibility measurement should be the first to show its own.

How it differs from the metrics it gets confused with

The two failure modes this table exposes are both common. A brand written about everywhere by third parties can be named often and cited never — good mention rate, zero citation share, and no control over how it is described. A publisher can be cited constantly and named never — its pages supply the facts while the recommendation goes to someone else.

Why citation share is the one to move first

Mention rate largely depends on what the rest of the web says about you, which you influence slowly. Citation share depends mostly on whether you have published a page that answers the question, which you control this week.

The citation data supports that. Across our eight questions in run A, 74 distinct domains produced 90 citations — 65 of those domains were cited exactly once, and the single most-cited domain accounted for five. Run B split the same 90 citations across 75 domains, 66 of them singletons. There is no small trusted club being consulted. The engine reaches for whichever page most directly answers the question in front of it, and plenty of what it reached for were companies’ own pages.

What counts as a good citation share?

There is no published benchmark worth quoting, and anyone offering you a single target number is guessing. The structural context is more useful than an invented target: across our eight questions in run A, 90 citations were spread over 74 distinct domains, so the average domain appeared 1.2 times (run B: 75 domains, the same 1.2). Against a field shaped like that, being cited even twice across a question set already puts you ahead of most of it.

Judge your own figure against your own previous month, on the same questions. That comparison is the only one that means anything.

How to raise it

1. Find the questions where you are cited zero times but your category is discussed. Those are the pages that do not exist yet.
2. Publish one page per question, titled as the question, answering it in the first two sentences. Not a category page with the answer in paragraph five.
3. Put the specifics in text on your own domain — prices, coverage, what you do and do not do. A fact in an image or a PDF cannot be cited.
4. Make the page worth returning to. A thin page retrieved once and found empty gives the engine a reason not to come back, and dilutes the pages that are good.
5. Re-measure monthly on the same questions and compare against your own last figure, not against a benchmark somebody invented.

One honest caveat about the number

Citation counts are noisy. We ran our audit twice, about a minute apart: the headline metrics reproduced exactly, but per-domain counts moved — ahrefs.com from 3 to 4, peec.ai from 4 to 2. Engines are non-deterministic, so treat a single run’s per-domain figures as estimates and the trend across months as the signal. Our full method, including how failed engine calls are handled so they never masquerade as absences, is on how we measure AI visibility.

Common questions