Skip to content
All insights
Measurement2 September 2026 · 6 min read

Being mentioned and being cited are different problems

Two outcomes that get reported as one number, with different causes and opposite fixes. Separating them is the first useful thing an AI visibility measurement can do.

Most AI visibility reporting collapses into a single figure: how often you show up. It is a reasonable place to start and a bad place to stop, because two quite different things are being counted.

A mention is your company name appearing in the text of an answer. A citation is a link to your domain presented as a source for that answer. They occur independently. You can be named in the prose with no link to you anywhere on the page, and you can be cited as a source for an answer that recommends somebody else.

Why the distinction has teeth

The two outcomes have different causes, so conflating them produces a strategy aimed at an average of two problems, which is to say aimed at neither.

  • Mentioned but not cited: the system knows about you, and it learned about you somewhere other than your website. The evidence is third-party. Changing your own pages will not necessarily change how you are framed, because the framing was written by someone else.
  • Cited but not mentioned: your content was useful enough to source, and not persuasive enough to be recommended. This is often a documentation or guide page doing the work while your product has no presence in the shortlist.
  • Neither: the interesting branch, because it splits again. Either you cannot be retrieved at all, or you can be retrieved and nothing about you is worth including.

The check that should come first

Before interpreting any of this, confirm the system can actually fetch your pages. A company blocked at the CDN and a company that is retrievable but unconvincing produce identical visibility numbers and require completely opposite responses.

What to record per execution

If you are building your own measurement, the minimum useful record per query execution is not a boolean. It is closer to this:

{
  query, surface, modelId, executedAt, repetition, geography,
  rawResponse,              // verbatim, never summarised in storage
  mentioned, position, matchedAlias, framing, excerpt,
  ownDomainCited,
  citedUrls[],              // and whether each one resolves
  competitorsNamed[]
}

The excerpt and the matched alias matter more than they look. Brand names that are also common words generate false positives at a rate that will quietly corrupt your numbers, and without the excerpt you have no way to audit a match after the fact.

Framing is the third variable

Once you separate mentions from citations, a third distinction appears almost immediately: how you were named. Recommended outright, listed among options, mentioned in passing, or named with a caveat attached are four quite different commercial outcomes that all count as one mention.

We record framing as a separate field for that reason. A company moving from 'mentioned in passing' to 'listed among the main options' has made real progress that a mention rate alone will not show.

What this changes in practice

For a company that is mentioned and never cited, the highest-value work usually runs through sources they do not own: review platforms, industry media, community threads, comparison content published by other people. For a company that is cited and never mentioned, the work is usually on their own site — the content is useful but the product has no presence in the parts of the site that answer buying questions.

Same headline number. Entirely different quarter of work. That is reason enough to stop reporting one figure.

See where you stand in AI search.

An AI Search Audit tells you how often AI systems name your company, who they name instead, and what is causing the gap. Every figure comes with the method behind it.