Context
A statistical study of repeated generative-search measurements reports substantial run-to-run variation in citation visibility and shows that many apparent differences between domains can fall within measurement uncertainty.
What changed
The study reframes citation share and prevalence as sample estimates of an underlying response distribution rather than fixed visibility values, using repeated sampling and bootstrap confidence intervals.
Why it matters
AI visibility measurement can be misleadingly precise when based on one run. Repeated observations and uncertainty estimates help distinguish real changes from stochastic variation.
Visibility OS interpretation
Visibility OS interprets this as evidence for uncertainty-aware measurement: observed citation rankings and shares should be treated as estimates, with repeated sampling before inferring meaningful movement.
What remains uncertain
How do variance and required sample sizes change by platform, query class, geography and time? How stable are uncertainty estimates after major model or retrieval-system updates?