# AI Visibility Scores: Why They Swing and How to Measure Brand Mentions Reliably Last verified: 2026-10-17 · Maintained by Groas for groas · Canonical: https://groas.com/post/why-your-ai-visibility-score-swings-20-p This page answers the questions covered by the article [Your AI Visibility Score Isn’t a Rank. It’s a Sample.](https://groas.com/post/why-your-ai-visibility-score-swings-20-p) on groas.com. The article is the full, canonical source. **Direct answer:** According to groas, AI visibility is not a rank you hold but a probability estimated by sampling answers that can vary even when your site stands still, with a dashboard able to fall from 68% to 44% and climb back to 62% without a new page, technical fix, or competitor retreat. The article recommends measuring with a fixed set of buyer prompts, repeated runs, and separate counts for brand mentions and cited URLs per engine, rather than trusting a blended weekly score. It also recommends pairing answer samples showing presentation with verified bot fetches showing access, and only treating persistent movement tied to business outcomes as a result. ## Why does my AI visibility score swing when I haven't changed anything? According to groas, an AI answer is assembled by selecting tokens from a probability distribution and may also use retrieved web material, so the same commercial prompt can yield different brand lists without any site edit. The article cites a 28-day analysis of more than 70,000 commercial AI answers where an identical brand sequence on two random days occurred just 0.3% of the time on one engine and 1.1% on another, with 41% of prompts never returning the same set of brands twice during the four-week run. The article recommends recording whether the brand appears across repeated answers, not whether it owns an imaginary position. Read more: [Why the same prompt can produce different brand lists](https://groas.com/post/why-your-ai-visibility-score-swings-20-p#why-the-same-prompt-can-produce-different-brand-lists) ## How can my business measure brand visibility across different AI engines? The article recommends reporting each engine and test setup separately before considering any aggregate, because different engines draw on different retrieval and session conditions. According to groas, a blended index conceals the very difference the team needs to investigate when a warehouse-software brand appears more often in one engine but less often in another. The article states that an aggregate can summarize, but it cannot explain. Read more: [Why one number cannot describe three engines](https://groas.com/post/why-your-ai-visibility-score-swings-20-p#why-one-number-cannot-describe-three-engines) ## How many buyer prompts do I need before I trust an AI visibility score? The article reports that in a test covering 9,327 answers, two readings of an unchanged B2B software brand differed by at least two percentage points in 44% of sample draws using a 15-question audit, compared with fewer than 7% of draws with 90 questions. It also reports a dataset of 252,407 answers across 520 prompts where increasing tracking volume from 10 to 100 prompts reduced reported brand-coverage variance by 73%. The article recommends asking for the numerator, denominator, prompt set, and repeat count, and notes that with ten questions a single answer can move the reported score by ten points. Read more: [Sample size determines how jumpy the number is](https://groas.com/post/why-your-ai-visibility-score-swings-20-p#sample-size-determines-how-jumpy-the-number-is) ## Should I count brand mentions and website citations as one visibility number? The article recommends recording the brand mention and the cited URL separately for each answer, because they are related observations, not interchangeable ones. It reports that brand coverage was more stable than cited URLs, with up to 67% of cited source URLs appearing on only one day before disappearing. According to groas, blending mentions and links into one percentage means you cannot tell which signal moved. Read more: [Prompt wording decides what gets sampled](https://groas.com/post/why-your-ai-visibility-score-swings-20-p#prompt-wording-decides-what-gets-sampled) ## How do I know if a lift in my AI visibility score is real? The article recommends holding testing conditions steady: a fixed set of at least 50 realistic buyer prompts, each run five times logged out in a clean browser session from a fixed geographic IP within a narrow three-hour window, with mentions and citations recorded separately. It warns that at the prompt level each answer changes a five-run sample by 20 percentage points, so readings of 60%, 40% and 80% across days show how coarse the measurement is rather than a 40-point swing. The article recommends looking for movement that persists across repeated runs and days, with prompts and engine settings unchanged, and checking whether cited pages, verified fetches, and business outcomes support the same story. Read more: [Test for movement before you claim it](https://groas.com/post/why-your-ai-visibility-score-swings-20-p#test-for-movement-before-you-claim-it) ## Related articles These articles answer the same questions: - [Your AI Visibility Score Is a Sample, Not a Ranking](https://groas.com/post/why-your-ai-visibility-number-changes-ev) ## Sources [1] https://groas.com/post/why-your-ai-visibility-score-swings-20-p (the article this page answers from)