Citation source

A citation source is a page that an AI actually retrieves and credits when it answers with web search turned on, and the set of citation sources determines whose version of the facts ends up in the answer.

In a retrieval-based answer, the model first fetches a handful of documents, then builds its answer from them. Those fetched documents are the citation sources. On platforms like Perplexity they appear as numbered footnotes; on others they show up as a source list or link cards; on some they are only visible if you expand the answer. Whether a brand can influence an answer depends to a large extent on whether its version of the facts is in that batch of documents.

What kinds of pages tend to get cited

• Pages that answer the question directly. Retrieval ranks by semantic relevance, so a page whose heading is the question and whose first sentence is the answer gets matched more often (see answer-first writing).
• Pages that can be crawled. A page whose body only appears after JavaScript runs, or that returns a blank page to crawlers, effectively doesn’t exist (see SSR vs. CSR and robots.txt).
• Pages with verifiable specifics. Content with complete numbers, dates, names and sources is adopted more readily than content built on adjectives.
• Sources that carry authority themselves. Trade media, official directories and encyclopedia entries usually rank ahead of personal content.

Your own content vs. third-party content

These two do different jobs, and you need both.

• Your own site answers “is there an authoritative version of this to check?”
• Third-party sources answer “is this claim credible?”

If only your own site makes the claim, the model may discount it as marketing material. If only third parties make it and your own site has no matching fact, versions start to diverge, which gives hallucination more room. See third-party evidence for more.

A common waste: third-party endorsements that only appear as text on your own website — “official partner of X,” “as featured in Y” — with no clickable link to the original. To a machine, a third-party endorsement without a URL does almost nothing. Link to the partner listing, the article, the award page.

How to find out what is being cited today

Run your question set on each AI search platform with web search on, and save every cited URL alongside the answer. Then group the URLs:

• your own pages (which ones, and are they the pages you’d want?);
• third-party pages that describe you correctly;
• third-party pages that describe you incorrectly, or describe a different company with your name;
• competitors’ pages and neutral roundups where you are absent.

That list is often the most actionable output of an AI visibility audit: it tells you which pages to fix, which to supplement and which publications to approach.

How it relates to the metrics

Progress on citation sources usually shows up first in mention rate and later in first recommendation rate, because being retrieved is the precondition for being recommended.

Keep in mind that citations only apply to web-connected answers. When a platform answers from the model’s internal knowledge alone, there are no citation sources to influence in the short term; what the model says reflects its training data. For the overall method, see generative engine optimization (GEO).