For AI brand answers, query language is the biggest source of answer noise (26.5%) while brand identity is tiny (1.5%), so repeating the same prompt is the least efficient way to spend a query budget.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Where Does the Noise Come From? A Variance-Components Decomposition of Non-Determinism in LLM Brand Answers
For AI brand answers, query language is the biggest source of answer noise (26.5%) while brand identity is tiny (1.5%), so repeating the same prompt is the least efficient way to spend a query budget.