Citation turnover measures how fast the pool of pages an AI engine cites changes across repeated observations. High turnover means the sources behind an engine's answers are largely replaced within weeks; low turnover means the same URLs keep getting cited.
It is a longitudinal, population-level metric, and that distinguishes it from its neighbours. Answer volatility is about one moment — the same prompt returning a different answer each time it is asked. Turnover is about time — aggregate across many prompts and weeks, what fraction of the cited set survives. A 2026 study (Digital Authority Partners) tracked 1,127 cited URLs in three waves over six weeks and found only about 10.6% appeared in all three, with roughly 40-60% of cited sources rotating monthly; it is single-vendor and directional, but the churn is large by any read.
Turnover is driven by three compounding forces: probabilistic sampling (answers vary from one sampling to the next), freshness weighting (retrieval favours recently-updated pages, displacing older ones), and a continuously changing index. For AI visibility, the implication is that a citation is a position you defend, not a win you bank: track citation coverage and share of voice as smoothed trends across many prompts and engines, and keep re-earning the citations that matter rather than assuming a single snapshot holds.