Back to Glossary
Data & Analyticsmeasurementnoisestatisticsai-visibilitymethodology

Noise Band

The pre-declared range within which fluctuations in AI visibility measurements count as chance and are not reported as change. Protection against over-interpretation.

Noise Band

AI answers are probabilistic: research from the University of St. Gallen (2026) shows that cited sources overlap only 32 to 43 percent when the same query is repeated on the same day. Whoever turns a single shift into a success story is reporting chance.

Why it matters

The noise band is declared before measuring: shifts of a few points count as noise; only movements beyond it, stable across several measurement waves, count as change. This protects both sides: the client gets no invented successes, the provider does not have to explain random dips.

The TYS view

The band comes with a frozen question set, several repetitions per engine, measurement waves across multiple days and a confounder log (model updates, seasonality). Single measurements are never reported as results.

Related Terms

First-Hand Source

Content that brings new, verifiable knowledge into the world from one's own observation, measurement or production, instead of rephrasing existing texts. The scarce resource of the AI era.

Hallucination Inventory

The systematic record of what AI systems say about a brand today: what is missing, what is wrong, and whose text the errors come from. The starting point of any serious visibility work.

Unified Object Graph

A connected knowledge graph in which products, services, documentation and policies live as objects with explicit relationships - website, API, llms.txt and MCP become renderings of the same knowledge base.

Have questions?

Our experts help you find the right digital strategy for your business.

Contact Us