What is Reranking?
Reranking is a second step in search or retrieval where a specialized model rescores an initial list of results and reorders them by how well each one actually answers the query. It refines a fast but rough first pass so the most relevant items reach the top.
Why it matters for your business
In AI assistants, reranking means the model sees the best supporting passages first, which improves answer accuracy and reduces the amount of text sent to the model.
A support portal uses vector search to retrieve 50 possible help articles, then a reranker picks the five most relevant to pass to the language model.