Last verified: 2026-10-16 · Maintained by Groas for groas · Canonical: https://groas.com/post/stop-counting-prompts-why-tracking-your
This page answers the questions covered by the article Stop Counting AI Brand Mentions. Fix the Pages That Never Get Cited. on groas.com. The article is the full, canonical source.
Direct answer: According to groas, small businesses should use prompts to find pages worth fixing, not to build an in-house mention-counting operation. The article argues that a name appearing in an answer is not a visit, lead, or revenue, while a citation a buyer can follow deserves attention. It recommends running ten to twenty high-intent buyer prompts as a periodic diagnostic, matching each omission to the URL that should answer it, and tracking whether canonical URLs earn citations and referral sessions that reach specifications, pricing, or a demonstration.
The article describes the familiar playbook as an API hook, a scheduled job, or a $99-to-$200-a-month dashboard that runs fifty variations on buyer queries, where a mention turns the cell green and an omission turns it red. According to groas, businesses should not build this into a weekly in-house operation. The article states the dashboard tells a content team that it was left out, not what to fix, such as why the brand appeared one day but vanished the next or which page needs changing.
Read more: The in-house tracker looks useful until you need an answer
The article states that no prompt-counting tool provides the diagnosis. A tracker records the appearance but cannot tell whether a page was never retrieved or was retrieved and passed over. It cites an analysis of 548,534 pages retrieved across 15,000 test prompts in which only 15% of retrieved candidate URLs were cited, and states that the mention count collapses both failures into one red cell.
Read more: The red cell cannot tell you why you were excluded
According to groas, it does not. The article states that synthetic mentions rising from 18% to 24% across fifty chosen queries still cannot point to the users, pipeline, or closed revenue behind that rise. It cites research putting clicks on cited AI URLs near 1.3%, versus roughly 29.2% for top organic results, and states that if an answer types a company name without linking to its site, that mention offers no direct click path.
Read more: A green cell is not a buyer
The article states that an AI answer is not a stable results page and retrieval context and model output can change without a change to a site. It cites research on non-determinism describing output changes even with temperature set to zero and greedy decoding enforced, plus live retrieval as another moving part. As a result, a brand could appear in eight of ten runs on Monday, two on Wednesday, and nine on Friday without changing a page, so a movement such as query #14 triggering a name in 42% of test runs is not by itself evidence that market position moved.
Read more: Mention share is impression share in a new costume
For every important omission, the article recommends identifying the URL on the site that ought to answer the buyer and putting it beside the sources the engine cited. The comparison is what each cited page makes easy to find, such as a direct diagnostic answer, a labelled specification table, a clear comparison, or public pricing detail. For a query about commercial freezer defrost-cycle troubleshooting, the article contrasts a cited guide with a short definition and accessible specification table against an answer inside a gated PDF or long promotional article, and recommends improving the page structure, publishing useful details where they can be crawled, and checking that the page is accessible.
Read more: Match each missed question to the page that should answer it
These articles answer the same questions:
[1] https://groas.com/post/stop-counting-prompts-why-tracking-your (the article this page answers from)