New research: AI visibility rankings are mostly statistical noise between runs
A paper covered by SEJ finds that brand visibility scores across LLMs vary significantly between identical prompt runs, making single-snapshot GEO dashboards misleading. The authors propose a stopping rule for when repeated sampling makes a ranking trustworthy. It's the first credible pushback on the wave of AI-visibility tools selling point-in-time scores.
If you're reporting AI visibility to clients or execs off one weekly pull, stop. Move to multi-run sampling with confidence intervals before the next board deck, or expect the numbers to swing and your credibility with them.