Hybrid search combines two different methods of searching: classic keyword search, which excels at exact names, codes and numbers, and vector search, which finds semantically related content even when the words used do not match at all. The results of both approaches are then combined and re-ranked. The reason is simple: vectors alone fail on unique strings such as a call's designation, a standard's number, or a product name, while keywords alone fail to capture a differently worded question. For content creation, this means that the old and new rules apply at the same time. Leave exact names, numbers, abbreviations and their spelled-out forms verbatim in the text, but at the same time cover the same topic in the user's everyday language too, including synonyms and question-form phrasing.
See also: Reranking, Content chunking, Query fan-out.
