Skip to main content
Citation share is the percentage of all source links in AI-generated answers that point to a given domain, across a defined set of questions. Where share of voice counts brand mentions, citation share counts URLs used as sources. It is the more actionable of the two. Mentions come from the model’s training data and move slowly; citations come from live retrieval and respond to content — see AI visibility.

​
How it is calculated

For a fixed query set, collect every cited URL from every answer, then:
Counting by domain rather than by answer avoids over-crediting a single answer that happens to cite four pages from one site.

​
What the distribution usually looks like

In our experience, across broad commercial query sets, cited domains fall into recognisable bands:
  • Official platform properties — the vendor’s own site for platform-worded questions, often a large share
  • Documentation and help centres — often a large share, because they are factual and directly answer specific questions
  • Primary specifications — standards bodies, search engine developer docs
  • Editorial and comparison sites — variable, often thinner than expected
  • A long tail — many domains at fractions of a percent each
The documentation band is the one a vendor can realistically enter, which is why help-centre content is the usual first move for improving citation share.

​
Interpreting the number

  • Volatility is normal. The cited URLs often change between two runs of the same query. Judge by rate across many queries over time.
  • Zero is a real starting point. A domain with no retrievable, factual pages will not appear at all, and no amount of query variation will surface it.
  • Query phrasing caps the ceiling. For prompts where one platform’s own domains take most citations, third-party citation share is structurally limited. Judge those prompts separately.
  • Citation share and conversions are not the same thing. A cited page still has to be a page worth landing on.