If you have ever tried to report on brand mentions, you have hit the definition problem. Is a mention a citation? Is an unlinked mention worth anything? Bradley Benner of Semantic Mastery gave the cleanest working answer I have heard, on the Unscripted SEO podcast.
A citation is a published mention of usually at least two out of four data points.
Bradley Benner, Unscripted SEO
The four data points are the classic local set – name, address, phone, website. What makes his framing useful is the threshold. One data point on its own is a namecheck. Two or more is something a machine can match against an entity it already knows about, which is when it starts doing work.
Why the threshold matters more than the label
Count every brand appearance as a citation and the number looks great and means nothing. Apply the threshold and you get a smaller number that tracks something real: how often the brand appears somewhere in a form that can actually be resolved to the business.
The precondition nobody reports on
Jason Barnard makes the point that sits underneath all of this – none of the credibility signals attach to anything until the machine knows who you are:
Because if it doesn’t understand who you are, it can’t attach credibility signals…
Jason Barnard, Unscripted SEO
Taken together the two are a sequence rather than two tactics. Be identifiable first. Then accumulate mentions that carry enough data points to be matched to you. Reversing the order is how brands end up with a pile of coverage that never consolidates into authority.
What to do with it
Audit your mentions against the two-data-point threshold before you report a number. Most dashboards will not do this for you, and the gap between the raw count and the qualifying count is usually large enough to change what you would recommend.
Related: what actually got people cited in AI search and becoming the cited source.
Both quotes verbatim from the recorded interviews, verified against the raw transcripts on 11 September 2026. Bradley’s fuller definition includes a term rendered ambiguously by the transcription and has been left out rather than guessed at.