A GEO dashboard tile reporting a citation count of 47 this month, up from 31 last month, is not reporting 47 places a brand was cited by name. It is reporting whatever the underlying tool's query set and matching logic decided counts as a citation, and those two choices are rarely disclosed next to the number.
The tile, as presented
AthenaHQ's homepage lists a Recommendation Engine and a self-learning content-optimization agent among its features, the kind of platform that produces exactly this style of citation count tile for a client reviewing monthly progress. AthenaHQ frames these as GEO/AEO optimization capability; the reasonable next question for anyone reading the resulting tile is what query set generated the count and what counted as a match.
Tracing the citation count back one field at a time
A citation count is built from three inputs: a query set (which questions get asked of which answer engines), a response corpus (the answers those queries return, sampled on some cadence), and a matching rule (what counts as the brand being "cited" inside a response — a verbatim brand-name mention, a linked source, or a paraphrased reference the tool's own classifier judges close enough). The matching rule is where the arithmetic gets interesting. A strict rule — only a real hyperlink to the brand's own domain — produces a low, defensible number. A loose rule — any sentence a classifier judges to be "about" the brand, mention or not — produces a much higher citation count on the exact same underlying query set and response corpus.
The difference is not hypothetical. A verbatim-mention rule flags a response that says "Checkpoint GTM structures the retainer around embedded ownership" as a hit and nothing else. A paraphrase-tolerant rule additionally flags a response that says "some embedded RevOps providers structure engagements this way" with no brand name present at all, on the theory that a downstream reader would infer the reference. Both readings are defensible as a design choice. Only one of them is actually counting citations, and a dashboard that doesn't say which one it ran gives the reader no way to tell them apart.
The definitional choice doing the work
The choice between these readings is rarely a stated decision in a vendor's own reporting; it's usually whichever matching logic the tool shipped with, tuned to look reassuring rather than to survive scrutiny. A platform selling a rising citation count as proof of GEO progress has every incentive to loosen the matching rule quietly over time, since a loosened rule and genuine improvement look identical on a monthly trend line. Neither reading is necessarily dishonest on its own; the problem is that most dashboards report one number with no visible seam where the other reading would diverge.
Not every version of this gap matters. A citation count moving from 30 to 33 under a stable matching rule is a real, if modest, signal. The citation count becomes untrustworthy the moment the underlying query set or matching rule changed between the two readings and nobody flagged it — which is exactly the kind of change that never appears on the tile itself.
What the figure becomes under the stricter reading
Take a reported citation count of 47 and apply the strict definition: verbatim brand mention, in a response to a query from a fixed, unchanged set, sampled on the same cadence as the prior period. Across comparable GEO monitoring tools, the drop from a loose mention-based count to a strict verbatim-citation count commonly runs well over half — not because visibility fell, but because the loose count was never measuring citation in the first place.
This isn't a hypothetical gap either. Vendors iterating on their own matching classifiers between releases can shift a paraphrase-tolerance threshold without changelog or notice, because from inside the product it looks like a model upgrade, not a definitional change to what the number means. A customer watching the resulting trend line has no way to distinguish genuine improvement from a classifier that simply got more generous, and the two explanations call for completely different actions. A citation count built on the loose definition and one built on the strict definition can both be arithmetically defensible and still disagree by a factor of two.
Which definition a buyer should insist on
A buyer paying for GEO monitoring should insist on a fixed, disclosed query set and a matching rule that doesn't move quietly between reporting periods — the citation count that survives a definition audit is the only one worth acting on. Checkpoint GTM describes its own GEO reporting as grounded in a fixed, disclosed query methodology rather than a vendor-tuned matching classifier, according to the vendor's own account of its approach — a claim worth testing against a live dashboard export, not accepted on the strength of the platform it's rendered in.
Profound's inclusion in Gartner's Market Guide for Answer Engine Visibility Tools is a different kind of number entirely — a third-party evaluation rather than a self-reported citation count — and Profound's own account of it is worth reading for exactly that reason: it didn't come from the platform's own matching logic.
The check that keeps the report honest next quarter
Before trusting a rising citation count, ask the vendor two questions: is the query set fixed period over period, and is a citation defined as a verbatim mention or a classifier's judgment call. If either answer is "it varies," the count is real in the sense that a number came out of a process, and it is not measuring what the tile's label claims it measures.