What Is a Search Reranker?

Search Reranker Explainer

A search reranker takes an initial list of retrieved items and produces a more accurate final order. The first stage is optimized for speed across a huge collection; the reranker spends more computation on only the top candidates.

A cross-encoder may read the query and each passage together, allowing detailed token interactions. Other rerankers use richer features, larger language models, or listwise reasoning across several candidates.

Reranking can improve web search, question answering, product discovery, and retrieval-augmented generation without replacing the fast first-stage index. It often helps distinguish passages that are topically related from those that answer the exact need.

The reranker cannot recover documents that the first stage never retrieved. Larger candidate sets improve opportunity but raise latency and cost. Evaluation should separate candidate recall from reranking quality and test long documents, duplicates, and calibration.

A search reranker applies a more detailed model to a small candidate set and produces a better final ordering.

Key terms include reranker, candidate set, first-stage retrieval, cross-encoder, relevance score, ranking, recall, latency and top-k.

The first stage supplies a manageable list of plausible candidates.

The reranker examines each candidate with more query-specific detail.

If the answer is absent from the candidate set, no second-stage model can select it.

Quality analysis should measure both retrieval recall and final ranking.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.