AI Search 101
What is an AI citation?
An AI citation is a source named or linked in a generated answer. It is not the same as a brand mention, where a company is described without any link. Citations can send traffic. Mentions shape what a buyer believes, and they are far more common.
The basics
| Also called | AI source, model citation, generative citation |
| Appears in | AI search results, AI Overviews, assistants such as ChatGPT and Perplexity |
| Distinct from | A brand mention, which names the company without linking to it |
| Depends on | Retrieval at query time rather than training data |
| Influenced through | Answer engine optimisation |
What is the difference between a citation and a brand mention?
This distinction decides how a company measures the channel, and most reporting gets it wrong.
A citation is a named source attached to an answer, usually as a link or a numbered reference. It is verifiable, it can be counted, and a share of readers click it.
A brand mention is the company appearing in the text of the answer with nothing to click. The model says a company does X, serves Y, and compares to Z, all from its trained impression. No link, no referral, no trace in analytics.
Mentions are far more common than citations, and for a business with a long sales cycle they matter more. A buyer who reads that a company is a credible option in its category has been influenced whether or not anything was clicked. The purchase, when it comes, will be attributed to a branded search weeks later.
How does a page become a citation?
Citations come from retrieval, not from training, and this is the practical opening. When an assistant fetches live pages to answer a question, the pages it fetches are the ones it can cite. So the question becomes what gets retrieved.
In practice the retrieved set favours pages that rank well for the question in conventional search, answer the specific question rather than a broad topic, state the answer plainly and early, load quickly, and are readable without executing much script. Pages behind logins, heavy interactive pages, and pages that bury the answer under a narrative tend not to survive the selection.
The uncomfortable corollary: a company can be described accurately and never cited, because the description came from training while the citations went to whoever answered the specific question best.
Which sources actually get cited?
Independent ones, disproportionately. Across categories the recurring citation sources are documentation, trade and technical publications, comparison and alternatives pages, community discussion such as Reddit, review platforms, and reference sites.
Vendor marketing pages are cited less often than vendors expect, for a reason that is easy to state and hard to accept: a system trying to describe a category treats a vendor's own claims about itself as weaker evidence than a third party's. This is the same instinct a careful human buyer applies.
Are AI citations worth pursuing?
Yes, but not as a traffic play. The clicks are real and small.
The value sits elsewhere. A citation is a durable signal that a page is the source a model reached for on that question, which tends to persist across sessions and across users. It also functions as third-party validation in front of a buyer at the moment they are forming a view of the category. Treating citations as a proxy for whether the reference content is doing its job is more useful than treating them as a traffic line.
How AI citations are used in marketing
Production. Reference pages built to be the best available answer to one question, with the answer stated in a liftable sentence near the top.
Analysis. Tracking which questions produce citations, which domains win them in a category, and whether a company appears as a citation, a mention, or not at all. The gap between mention rate and citation rate is usually the most informative number.
Distribution. Working on the independent sources that get cited, since a mention in a publication a model trusts often outperforms another page on the company's own domain.
How this impacts your business
A language model does not look a company up in a register. It composes a description from everything it read during training and whatever it retrieves at the moment of the question. Trade press, directories, forums, competitor comparison pages, and somewhere in that mix the company's own website too.
So the description a buyer sees is assembled, not published. No company controls it, but every company supplies a share of the material it is built from.
The traffic impact is still small, roughly one percent of sessions for most companies. The influence is not small, because the reader is often a buyer forming a first impression of what a company does and who it serves, before any page gets a click.
Checking takes ten seconds. Type the company name into ChatGPT and read the answer.
Frequently asked questions
Do AI citations pass link equity?
No. They are not links in the search engine sense and carry no ranking value. Their worth is exposure and credibility rather than authority transfer.
Why does the same question produce different citations each time?
Retrieval is not deterministic. The retrieved set varies with phrasing, session context, model version and timing, so citations shift between runs. This is why single checks are unreliable and tracking over time is not.
Can a company pay to be cited?
Not in the answer itself. Some assistants carry advertising in separate placements, but the cited sources are selected by the retrieval system.
Is a mention without a link worth anything?
Usually more than the link. It reaches a buyer at the point they are deciding what the category contains and who belongs in it.
How do I track AI citations?
Monitoring tools run a set of representative questions across models on a schedule and record which sources appear. Manual checks work for spot verification but not for measuring change.
Related definitions
- What is AI search?
- What is answer engine optimisation?
- What are AI Overviews?
- What is a large language model?
- AI Search 101
Written by Lari Numminen, Generate More. Also available in Finnish. Updated 31 August 2026.