LLMInference & InterfacesUpdated 2026.04.28

Reranker

Also known as재정렬 모델Cross-encoder후보 재정렬

In one line

A reranker re-scores the candidate documents that retrieval produced and reorders them — the final gate that decides which sources an AI actually cites.

Going deeper

A reranker re-scores the top N candidates retrieval produced, feeding each one back through a more precise (usually cross-encoder) model alongside the query. It costs more, but accuracy steps up noticeably. Cohere Rerank, BGE Reranker and bespoke trained models are common picks.

For marketers, the key insight is that getting retrieved is not the same as getting cited. A typical AI answer cites 3–5 sources; the reranker decides who clears the final cut. Matching query intent precisely matters as much as showing up at all.

Practitioners consistently rank rerankers as one of the highest-leverage components in a RAG stack. When AI answers feel off, the reranker is usually the first place to look — and GEO diagnostics often surface the exact pattern of 'retrieved but not cited'.

Related terms

How does your brand show up in AI answers?

Villion measures how your brand appears across ChatGPT, Perplexity and AI Overviews, then automates the work that lifts citation rate and share of voice.

Get a free audit