What is Rerank survival rate?
Rerank survival rate is the fraction of brand chunks in the inferred candidate set per priority sub-query that survive the rerank pass and appear in the cited synthesis surface across a rolling 4-week window. It is the headline metric of a rerank-optimization program and the chunk-level analog of citation share. Compute per sub-query (cited brand chunks divided by inferred candidate-set brand chunks) and average across the priority sub-query set. A rate above 25% is category-leading; 15–25% is competitive; below 15% is exposed. Mid-2026 cohort medians sit at roughly 12% on mid-market programs and 22% on category-leading programs. The rate is the single best predictor of citation share lift per editorial hour at constant retrieval coverage — chunks that retrieve and rerank actually appear in the answer; chunks that retrieve but fail rerank do not.
How it relates to AI UGC
Multimodal-active sub-queries carry a parallel visual rerank survival rate — image chunks scored on visual cross-encoder properties (ImageObject schema density, persona stability, image freshness, caption alignment, content-hash recency). Programs tracking the parallel rate identify visual rerank decay 5–7 weeks ahead of programs tracking text rerank alone. ppl.studio production cadence holds the visual rerank survival rate above the carousel threshold on persona-locked image sets.
Key statistics
- Mid-2026 rerank survival rate medians: 12% on mid-market programs, 22% on category-leading programs; rates above 25% are category-leading (survival-rate cohort, 2026).
- Rerank survival rate and total citation share move together with a 4–8 week lag once chunk-level edits start shipping — the lag is the engine's re-embedding cycle for re-scored chunks (survival-lift lag audits, 2026).
- Programs that lift rerank survival from 12% to 22% on the priority sub-query set report total citation share lift of 1.8–2.6× at constant retrieval coverage — the lift isolates rerank optimization from retrieval optimization on the metric line (rerank-lift attribution, 2026).