Skip to main content
A mention is your name in the answer text with no link attached. A citation is one URL the model attached. A source is a domain bucket across many citations. Three different objects, three different counts — and confusing them is the single most common misreading of AI visibility data.

​
Mention, citation, source: what is the difference?

In the engine’s words: “Mentioned is not cited: a model can name a brand it never links.” The reverse also happens — a model can cite a page on your domain in an answer that never names you, because the page was useful evidence for something else.
The nesting is straightforward. Many citations roll up into one source. Many runs roll up into one share of voice or citation rate. Rates count runs; sources count URLs. Never compare a rate to a source share and expect them to reconcile — they have different denominators and different units.

​
What is source share?

Source share is a domain’s slice of all citations in the window: citations to that domain, divided by every citation Attensira recorded across all your tracked runs in that window. Unlike mention and citation rates, source shares do describe parts of a whole, because each citation belongs to exactly one domain.
That makes source share a map of the citation landscape around your prompts: which publications, review sites, forums, documentation hubs and competitors the models actually draw on. Your own domain appears in this list like any other. A high source share for a review site means models lean on it when answering your prompts — which tells you where being present is worth the effort. Source share is sensitive to which prompts and which models you track, since those choose the runs the citations came from. Widen your prompt set and the source mix will shift.

​
What does average citation position mean?

Average citation position is the mean index of a URL within the model’s citation list, calculated per source domain. If a domain’s URLs tend to appear as the model’s first attached link, its average position sits near the top of the list; if they show up after several others, it sits lower.
Average position is an index into the citation list. It is not a rank of your brand, not a search ranking, and not a competitive leaderboard position. Where your brand came in the answer’s text is a different number, counted among the names you track; see Share of voice.
Read it as a weak ordering signal and nothing stronger. Models order citations for reasons that are not documented and not stable, and an answer with two citations produces very different index arithmetic than an answer with twelve. Position moves are worth noticing only alongside a change in source share or citation rate; on its own, a shift from 2.4 to 2.1 is not evidence of anything. A citation says the answer leaned on a page. It does not say what for. So every citation carries its context: the sentence of the answer that carries the link, verbatim. “Acme is the easiest to set up [2]” beside the review it cites tells you which claim that page is propping up, and that is the claim you would have to answer on that page or elsewhere. The context is found in the answer’s own text, by the link itself, then by its footnote marker ([2], [^2], 【2】), then by its domain written in the text. The answer’s sentences are searched first and its reference lines (a footnote definition, a sources list) only after, because a sources list names the page without saying why it was cited. In a list item or a table row the context is that line. A context is at most 300 characters; a longer sentence is cut to a window around the link, on a word boundary. The context is never invented. It is always a substring of the answer, and when the text carries no trace of a link, the context is empty: Contexts are kept for answers from the last 90 days: the stored answers in that window were read again when contexts were introduced, and every new answer gets them as it is read. An older answer shows the dash, not an empty context, because an empty context would claim we looked. On the prompt page, pick a day and an engine on the day-by-day strip to open that day’s answers: each lists its links with their context under Links and what they were cited for, beside its brand order. Over MCP and the REST API, each citation on answer_history and answer_slice rows carries context beside url, domain and position.

​
How to read these numbers

Like every other number in the product, source and position figures carry sample information. See Reading the numbers for how to handle small samples and unmeasured cells.