Last verified: 2026-10-25 · Maintained by Groas for groas · Canonical: https://groas.com/post/7-myths-about-measuring-your-brand-s-ai
This page answers the questions covered by the article 7 Myths That Make AI Visibility Reports Look More Certain Than They Are on groas.com. The article is the full, canonical source.
Direct answer: According to groas, a single sampled AI answer is one answer at one moment to one prompt and does not prove a durable ranking. The article recommends measuring how often the brand appears across repeated answers to a fixed prompt panel, for example 22 mentions in 50 runs as a 44% observed share. It also recommends logging footnote citations, brand mentions without links, and primary recommendations separately and comparing the 30-day moving average against a baseline with every content change logged.
According to groas, a small business can build a useful baseline by choosing 20 to 50 conversational buyer prompts from closed-lost reasons, onboarding conversations, and paid-search term reports and keeping the panel fixed for at least 90 days. The article recommends running the prompts repeatedly three times a week with consistent settings, logging the setting if a temperature between 0.2 and 0.5 is used, and recording separately whether the brand was named, the domain was cited as a linked source, and the product was recommended. The article recommends comparing the 30-day moving average share of answers to the baseline, for example 12% of sampled runs in month one versus 28% in month two as a change worth investigating, and logging every change with its timestamp.
Read more: Myth 7: “Monitoring Is the Work”
The article states that one run cannot tell whether an answer favors the brand or was a convenient variation, because language-model answers can vary between runs even when the prompt stays the same. It recommends 3 to 5 repeated runs per prompt across a fixed panel of 50 to 300 buyer queries and repeating the panel over seven days to see whether a mention is common or unusual. The article positions one prompt run as a screenshot and repeated measurement as the starting point for comparing one period with the next.
Read more: Myth 2: “I Checked the Prompt, So I Know Where We Stand”
According to groas, a fetch is not a citation and counting crawler requests alone can tell the wrong story. The article cites a study of 308.6 million enterprise server-log hits where citation visibility fell on major sites during the period examined while search-bot crawl traffic to those domains tripled. It states that training crawls, search-indexing crawls, and on-demand fetches in response to a user request are different activities, so a logged crawl should not be recorded as a recommendation that was not seen.
Read more: Myth 4: “An AI Bot Fetched Our Page, So We Got Cited”
The article recommends logging three outcomes separately because a blended score conceals commercially different events. A footnote citation means the page supplied reference material but the answer did not endorse the product, an entity mention means the brand appeared without a linked source, and a primary recommendation means the answer presented the product as a leading fit for the buyer’s stated need. The article states the third is closest to the decision and the three should not receive the same weight in a commercial report.
Read more: Myth 6: “A Mention Is a Citation Is a Recommendation”
[1] https://groas.com/post/7-myths-about-measuring-your-brand-s-ai (the article this page answers from)