# Measuring Brand Visibility in AI Search Without Single-Run Scores

Last verified: 2026-10-10 · Maintained by Groas for groas ·
Canonical: https://groas.com/post/your-ai-visibility-score-is-a-coin-flip

This page answers the questions covered by the article [Your AI Visibility Score Is One Run Dressed Up as a Trend](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip) on groas.com. The article is the full, canonical source.

**Direct answer:** According to groas, a single run of prompts is not a visibility measurement because within-prompt resampling alone accounted for 34.8% of output variance, even at a low decoding temperature. The article recommends building a panel of 20 to 30 unbranded buyer prompts, running each prompt 3 to 5 times per engine, and tracking recommendation frequency separately by engine and prompt group. It advises treating a shift of less than 15 percentage points, or one that does not persist across two monthly testing windows, as insufficient grounds for a budget move.

## How can my business measure brand visibility across ChatGPT, Gemini and Google AI Overviews?

The article recommends building a panel of 20 to 30 unbranded buyer prompts covering problem-aware, category-comparison and high-intent constrained questions, with the brand name removed from every query. For each engine tracked, it advises running every prompt 3 to 5 times within a tight testing window and logging engine, exact phrasing, run number and full answer. Results should be kept separate by engine rather than blended, with Recommendation Frequency calculated as runs that recommended you divided by total runs and Citation Rate tracked separately.

Read more: [Hold the setup steady, then repeat the runs](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip#hold-the-setup-steady-then-repeat-the-runs)

## How can a small business automate brand monitoring in AI search in-house?

The article describes a one-spreadsheet, two-hour monthly audit that tracks date, exact prompt phrasing, buyer intent stage, engine, run number, recommendation status of Recommended, Mentioned or Absent, and footnote URL if present. According to groas, on the first Tuesday of each month a small team should run the same 25 core prompts through each chosen engine five times, which requires roughly 125 calls per engine. For each prompt and engine, the article directs calculating Recommendation Frequency as runs that recommended you divided by total runs, and calculating Citation Rate separately.

Read more: [The in-house version: one spreadsheet, one afternoon](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip#the-in-house-version-one-spreadsheet-one-afternoon)

## Why does my AI visibility score change week to week when I changed nothing?

According to groas, an LLM answer is generated token by token rather than retrieved from a stable index, so the answer can change even when the prompt does not. The article reports that only 37% of featured brands remained consistent across repeated runs under identical prompts, with simple factual queries diverging by 5% to 10% and complex recommendation and evaluation queries showing 40% to 60% output divergence. It recommends repeating the identical prompt before calling a change a trend, because a weekly drop from 55% to 30% may reflect normal movement rather than degraded standing.

Read more: [The answer changes even when the prompt does not](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip#the-answer-changes-even-when-the-prompt-does-not)

## Why is a single blended AI visibility score misleading?

The article states that rolling multiple AI interfaces into one composite buries which interface actually produced the mention. According to groas, in an analysis of 300 business software queries, broad queries showed 70% recommendation variance across platforms, and Google AI Overviews showed greater run-to-run volatility than the standalone chat models. The article recommends keeping results separate by engine and reporting visibility by buyer problem, not just by category.

Read more: [A blended engine score hides the useful result](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip#a-blended-engine-score-hides-the-useful-result)

## How do I tell if an AI mention is a real recommendation or just a citation?

The article recommends recording for every answer whether the brand was recommended as a vendor solution, mentioned without a recommendation, cited through a linked source, or absent, keeping the citation field separate. According to groas, in an audit of 1,796 commercial category queries, a domain was cited 142 times but recommended as a vendor zero times, and citation of technical content has coincided with recommendation of a competitor. The article states that a source citation is not a buying recommendation and advises counting both but not selling one to leadership as the other.

Read more: [Measure what the model actually did](https://groas.com/post/your-ai-visibility-score-is-a-coin-flip#measure-what-the-model-actually-did)

## Sources

[1] https://groas.com/post/your-ai-visibility-score-is-a-coin-flip (the article this page answers from)