Put the sample size next to the change rate
VisibilityOS saw brand presence change in 82 of 817 matched AI answer comparisons. The overall rate was 10.0%. Here’s how the counts and rates varied across the eight engine views.
What the study observed
An answer counted as present if it named the brand or linked to its official site. A pair changed when that combined label flipped.
| Engine view | Changed / matched | Rate |
|---|---|---|
| Grok | 21 / 81 | 25.9% |
| Gemini web | 15 / 106 | 14.2% |
| Claude | 14 / 107 | 13.1% |
| Gemini | 9 / 107 | 8.4% |
| Google AI Mode | 7 / 95 | 7.4% |
| Perplexity | 7 / 107 | 6.5% |
| ChatGPT | 6 / 107 | 5.6% |
| ChatGPT web | 3 / 107 | 2.8% |
Rates describe the questions, settings, and dates in the study. The same questions were used more than once.
Same 10% rate, different precision
Three fictional samples show how much the range narrows when you collect more independent comparisons.
Each marker is the observed 10% rate. The bars use an independent-comparisons model for the fictional examples. Repeated questions in the study can produce correlated results.
What to report in your next audit
Report both the count and the rate for each engine and setting. Keep question text, model, search settings, language, and market aligned between rounds. When the same questions repeat across days, keep the question IDs so you can account for repeated observations in later analysis.