A citation is a source link an AI answer attaches to a claim. It is the page the model retrieved and leaned on when composing its response.
Citation-first engines like Perplexity attach sources to every answer; ChatGPT and Gemini cite when they browse. Citations are the paper trail of AI recommendations: the pages a model cites decide the brands it names.
For brands, citations are the most actionable layer of GEO. You can see which review sites, comparison pages, and community threads the models actually read for your category. Then make sure your brand is represented accurately on them.
Why it matters now
In a grounded AI answer, the brands that get named are downstream of the pages the model retrieved. Change the sources and you change the recommendation. That makes citations the most actionable layer of AI visibility: unlike a model's trained memory, the pages it reads at answer time are something a brand can influence.
It matters now because citation-first engines are becoming the default for buying questions. Perplexity attaches sources to every answer; ChatGPT and Gemini cite when they browse. The cited pages are a public map of what the models trust in your category, and that map is one a brand can actually read and act on.
How WhereDoIRank measures it
When we record the answers at each close from ChatGPT, Claude, Gemini, and Perplexity, we keep the citations attached to each response as part of the receipt. So a reported ranking is not just a number: it comes with the pages the model leaned on to produce it.
Reading those citations across a category shows which review sites, comparison pages, and community threads the models actually consult, and whether a brand is present and accurate on them. That is the difference between guessing at what to fix and seeing the sources the recommendation was built from.
A worked example
Say a brand is absent from Perplexity's answer to a category question. The citations show the engine leaned on two comparison articles and a "best of" roundup. The brand appears in none of them. The gap is now concrete and addressable: get listed, accurately, on the pages the engine reads.
After the brand is added to those sources with correct details, a later run can retrieve them and fold the brand into the answer. The citation trail turned an opaque absence into a specific to-do list.
Common misconceptions
A citation is not the same as a recommendation. A page can be cited as a source without the brand on it being the one the answer recommends, and a brand can be recommended while its own site goes uncited. Citations are also not a ranking hack: cramming links onto a page does not earn them, because the model cites what it retrieved and found useful, not what a page asks to be cited. And they are not permanent; as the underlying pages change, so do the citations, which is one reason grounded answers shift week to week.
More citations are not automatically better, either. A single accurate, well-regarded source the models trust can carry more weight than a dozen thin mentions. What matters is being present, and correct, on the specific pages an engine reaches for in your category, not blanket coverage.
How it connects
Citations are grounding made visible: the retrieved sources an answer stands on. They are the raw material of AEO, where the goal is to be the cited source. And they live inside receipts, the stored answers, with their sources, that let a reported rank be checked against what the model actually read.