Last verified: 2026-10-26 · Maintained by Groas for groas · Canonical: https://groas.com/post/the-prompt-you-track-isn-t-the-search-th
This page answers the questions covered by the article Your AI Visibility Tracker Sees the Prompt, Not the Searches Behind It on groas.com. The article is the full, canonical source.
Direct answer: According to groas, an AI visibility tracker replays the buyer prompt and counts brand mentions, but the assistant can rewrite one prompt into several hidden sub-searches and assemble an answer from those retrievals. The article compares the prompt to a keyword and the hidden searches to a search-terms report, so a visibility percentage shows the outcome without showing which search produced it. The article recommends diagnosing whether brand material was retrieved for a relevant sub-query, whether it was cited, and whether the brand was recommended, then fixing the missing passage for that specific need.
According to the article, Tuesday’s rewrite may have found a pricing passage while Thursday’s leaned toward a fleet-safety question the page does not answer, so a Tuesday mention followed by a Thursday absence does not, by itself, mean content deteriorated overnight. The article cites analysis finding only 27% of fan-out queries repeated across runs of the exact same prompt, while 66% appeared once. It also reports pages ranking for both a primary query and its fan-out sub-queries were 161% more likely to be cited than pages ranking for the primary query alone.
Read more: Step 2: each sub-query creates a separate retrieval opportunity
The article states one prompt can open several different routes to an answer, citing analysis of over 20 million fan-outs that reported an average of 2.3 to 2.8 internal web searches per prompt with average sub-query length growing from 6 to 12 words as rewrites became more specific. For the example question about the best dispatch software for a 15-van medical courier business, the assistant could pursue separate jobs such as compliance requirements, fleet-size fit for fleets under 20 vehicles, and reviews and pricing. The article advises identifying the distinct jobs the answer has to perform rather than writing one page for every possible rewrite.
Read more: Step 1: one buyer question becomes several narrower jobs
According to groas, finding a page is not the same as recommending a product, and being retrieved, cited, and recommended are distinct events even when a tracker compresses them into one visibility score. The article gives three diagnostic questions in order: did material about the brand appear in a result for a relevant sub-query, did the assistant use or cite that material when explaining the answer, and did it name the brand as a vendor that fits the buyer’s need. The article states a “no” at each stage calls for different work, and more prompt replays will not tell which stage failed.
Read more: Step 3: retrieval, citation, and recommendation split apart
The article states if a platform only gives an aggregate percentage for twenty fixed prompts, it does not tell what to fix. According to groas, a useful tool should show sub-query intent grouped by distinct questions, distinguish material that was fetched from a URL that was cited and a brand that was explicitly recommended, and point to the specific content gap such as an absent pricing boundary or unaddressed compliance criterion. The article describes this as the difference between a dashboard that says a number fell and an operating instrument that gives a plausible place to look.
Read more: Demand a diagnosis, not another visibility percentage
[1] https://groas.com/post/the-prompt-you-track-isn-t-the-search-th (the article this page answers from)