engines
Retrieval engine
The component of an AI answer engine that decides which web pages to read before generating an answer.
Definition
A retrieval engine is the subsystem inside an answer engine that, given a user query, fetches the top-K candidate documents to feed the language model. Modern retrieval engines use a hybrid ensemble: dense embeddings (semantic similarity), sparse BM25 (lexical match), domain authority weighting, and recency boost. The citation decision is made HERE, not by the language model.
Example
Perplexity's retrieval engine scores each candidate URL using α·cosine + β·BM25 + γ·authority + δ·recency, then passes the top 5 to its synthesis model.How to optimize
Score high on all four ensemble axes: semantic (use buyer vocabulary), lexical (cover query terms), authority (Wikidata + outbound citations), recency (fresh dates). VectorCite's retrieval simulator scores all four.