Reading a visibility score: trend over number

A visibility score may drop from 34% to 28% without proving that marketing performance has worsened. Before you conclude anything, look at the sample size, the tracked questions, and the uncertainty attached to each estimate.

AI answers are samples, not facts

If you pose the same buying question to the same assistant five times, you should not expect five matching brand lists. Models draw samples; retrieval changes; the roundup article referenced at 9am might differ from the one cited at 4pm. In that sense, a visibility score is an estimate based on repeated sampling: out of the times buyers asked those questions, how often did your brand appear?

The practical implication is immediate: any one execution, even a quick manual “let me just check ChatGPT,” is just a single draw from a distribution. One run cannot reveal your visibility any more than one coin toss can reveal whether a coin is biased.

What the confidence band is telling you

A credible visibility score includes a band, such as “34%, ±8.” More runs make the band tighter; fewer runs make it broader. That band is the system’s way of showing how much evidence supports the number. Two basic guidelines:

• Be cautious with minor shifts. A move from 34% to 30% with overlapping bands does not, on its own, demonstrate a decline. Keep the observations and look over a longer stretch.
• Look into persistent changes. If bands separate, or the direction repeats across several readings, it may warrant closer review. Still, neither factor by itself proves what produced the shift.

Trend beats level; share beats trend

Your score’s absolute value is influenced by variables outside your control: how list-like your category’s answers are, and how many brands can fit into a single result. Comparing your 30% with another industry’s 60% tells you nothing useful. Two comparisons are meaningful:

• You compared with you, over time. Using a stable prompt set makes the observations more comparable. A score that changes after a source edit does not, by itself, show the edit caused the change.
• You compared with competitors, using the same questions on the same days. This keeps both the question set and the observation dates fixed. Overtaking a competitor is a visibility outcome; check traffic and sales separately before concluding anything about revenue.

Judging whether your fix worked

The discipline that makes visibility tracking worth the cost is this: when you launch a change, whether a comparison page, a review campaign, or a roundup placement, record the baseline first and assess only the runs that come after the change. Then allow enough runs to accumulate. Drawing conclusions from two days of data is how teams end up believing nothing works. Give the change a few weeks and a dozen runs before deciding, and expect different platforms to update at different speeds: live-browsing surfaces can shift in days, while model-memory surfaces may take months.

Geofound includes action and outcome tracking so you can review observations before and after your work. Its SEO and paid ads research adds context: competitor rankings, keyword opportunities, backlinks and observed search ads. Keep those measures separate from your AI visibility score. A competitor ranking for a term does not prove it appears in an AI answer, and an ad does not reveal the advertiser's sales.

If you still lack a baseline, the free scan gives you three buyer questions and the brands ChatGPT names. The plans include ongoing AI tracking, SEO and paid ads research, and practical recommendations.