Citation Share is the share of the sources AI assistants cite — when answering the kinds of questions your customers ask — that you control, versus sources your competitors control, versus neutral third parties like directories and review platforms. Of the three numbers we track, it’s the least famous and the most useful early, because it usually moves before the answers themselves do. If Discovery Rate tells you whether you’re winning, Citation Share tells you whether the ground is shifting — and it’s the first place a fix shows up.
Here’s what the metric is, why it leads, and what actually moves it.
Every grounded answer shows its reading list
The assistants your customers use — ChatGPT, Google Gemini, Perplexity, and the rest — are grounded: they search the web before answering and build the response from what they just read. Because the answer comes from retrieved documents, it arrives with citations attached — the specific sources the assistant read on its way to the reply.
Those citations are receipts. They tell you, for every answer, whose material the assistant trusted enough to build on. Collect them across many questions and many runs, and a pattern emerges that no single answer shows: for your kind of question, in your market, whose story are the assistants reading?
Citation Share is that pattern, made into a number.
The three buckets
Each cited source lands in one of three buckets:
- Yours — sources you control: your website, your Google Business Profile, listings and profiles you maintain. When these are cited, the assistant is reading your version of the facts.
- Your competitors’ — their websites and profiles. When these carry the answer, assistants are learning about your market from someone else’s material, and the recommendation tends to follow the reading.
- Neutral — directories, review platforms, map services, local press. Nobody owns these outright, but their entries about you can be right or wrong, claimed or abandoned.
The distribution is the diagnosis. If competitor-controlled sources are carrying the answers, you can see exactly which pages are winning and what they say that your material doesn’t. If neutral aggregators are carrying them, the fight is for the accuracy of your entries on those platforms — which is very fixable. And when a competitor gets the recommendation, the citations turn the loss into a to-do list: you don’t just learn that they got the call, you learn which source earned it for them, and whether you can be on it too.
Why it moves first
The causal chain of AI visibility runs one direction: sources → citations → answers. You can’t edit an assistant’s answer. You can only edit what it reads — and the reading changes before the answer does.
Walk one fix through the chain. Say your profile on a review platform holds the wrong hours, and answers about your business keep hedging or skipping you. You correct the entry. The corrected page exists immediately; retrieval starts encountering it as the platform’s systems re-read the web. On the next test cycle, the citations shift — the corrected source starts appearing under answers, or your own site does, where before a stale directory or a competitor’s page carried the load. The wording of the answers may not have changed yet. The inputs have, and Citation Share is the number watching the inputs.
Being named in unaided answers — Discovery Rate — sits at the far end of that chain, behind an assistant’s judgment about confidence and consistency that nobody can script. It’s the number that matters most and moves last. Citation Share is the chain’s midpoint: close enough to the fixes to respond quickly, close enough to the answers to mean something.
That’s also why we lean on it for honesty’s sake. In the early months of source work, “the answers haven’t changed yet” and “nothing is happening” are different claims — and Citation Share is how you tell them apart with evidence instead of reassurance.
Where it sits among the three numbers
We report three metrics, deliberately separate, because they answer different questions at different speeds:
- Discovery Rate — when customers ask without naming you, how often do you appear? The outcome. Slowest to move.
- Answer Accuracy — when assistants talk about you, is each fact correct, wrong, or missing? The quality check on what’s being said.
- Citation Share — whose sources are being read? The leading indicator, and the earliest feedback that fixes are landing.
One number would be tidier, and it would lie by averaging. A business can be visible but described wrong, or accurately described but rarely found — and a single blended score hides exactly the distinction that tells you what to fix next.
What moves it
Citation Share responds to the fixable-sources list — the same unglamorous inventory that underlies everything in AI visibility:
- Your Google Business Profile, complete and current — it feeds more answers than any other single source.
- One identity, not two. Duplicate listings split your citations the way they split your reviews, and a contradicted source is a source assistants learn to route around.
- Your website’s facts, machine-readable — schema markup makes your own pages easier to cite for the facts they state.
- Pages that answer real questions. If nobody’s page answers “do they take walk-ins,” the assistant improvises from whatever’s adjacent. Publishing the answer gives retrieval something of yours to find — and cite.
- Your entries on the neutral platforms — directories and review sites cited in your market, checked for accuracy and claimed where possible.
Measured honestly, or not usefully
Two cautions from our own practice. First, assistants vary their answers, so citations vary too — a source cited in one run may not appear in the next. A Citation Share worth acting on comes from repeated runs of each question with the variation reported, not from one afternoon’s screenshots. Second, the bucket definitions have to be documented: which domains count as yours, which as competitors’, which as neutral. A metric with movable definitions can be made to say anything, which is the same as saying nothing.
You can watch the raw material yourself today: our free DIY walkthrough includes reading the citations under your own answers — most owners are surprised by whose pages are doing the talking. And if you’d rather have the whole chain measured properly — repetitions, documented buckets, confidence intervals, every claim traceable to a dated transcript — that’s the service, and the first look is free.