Updated 6 Aug 2026
Precision comes from total observations at the level you report, not from a fixed number of runs per prompt. The margin of error shrinks with the square root of the sample size.
Approximate 95% margin of error at the worst case (a rate near 50%):
Rates far from 50% tighten faster. A brand at 95% presence needs fewer runs to pin down than one at 50%.
Halving the margin of error quadruples the sample. Going from ±10 to ±5 takes roughly 4× more observations.
There is no upper limit. The range shrinks toward zero forever and never arrives. “Enough” is the precision at which a tighter answer would not change the decision.
Drift caps useful volume. Past a few hundred samples per cell in a single window, you are averaging a moving target. The fix is a tighter time window, not more runs.
Gross presence or absence
±15–20 points. Roughly 25–40 observations. Tells you “showing up” from “invisible.”
A defensible headline rate
±10 points. Roughly 96 observations. Enough for a competitive comparison.
A “we moved the needle” claim
±5 points or tighter. Roughly 385+ observations. Two bands must stop overlapping before a change is called real.
An initial read is sized to a minimum of 100 observations per reported cell, per platform, per market. At a rate near 50%, that places the 95% range at roughly ten points at the worst case, and considerably tighter for rates further from the middle. It is enough to establish a defensible headline rate and a variance floor that later readings can be measured against.
A cell is the level at which a number is reported. Presence on ChatGPT in the United States in English is one cell. The same brand on Claude in France in French is another. We do not pool cells to reach a sample size, because pooling is what produces a global average that hides the variation you need to decide on.
Where a decision requires a tighter answer, for example establishing that a change between two reads is real, the sample is sized up for that specific comparison. We say in the report which figures are sized for a headline read and which are sized to support a comparison.
AIVO is an independent AI brand representation firm. We measure how AI platforms describe a brand across presence, comparison, recommendation, accuracy, sentiment, sources, access, and audiences. Every rate is reported as a range at 95% confidence, per platform and per market, and every finding traces back to the prompts, responses, and citations behind it. We do not sell content, PR, technical remediation, or any other implementation, and we take no referral fee from the partners we recommend. The AI Brand Representation Report starts at $4,000 for one market, one language, and one core offering, delivered in 7 to 14 business days.
Independent measurement of how AI represents your brand. One market, one language, delivered in 7 to 14 business days.
30-minute scoping call · no commitment