What is hybrid search?
Hybrid search runs a keyword search and a vector (semantic) search over the same data at the same time, then merges the two result sets into one ranking. It exists because the two methods fail in opposite ways: keyword search misses anything phrased differently, and semantic search drifts away from exact terms like a model number or a postcode. Running both and combining them recovers most of what either would have missed on its own.
Keyword search: exact, and brittle
Keyword search matches the words you typed. It is precise about identifiers — a VIN, a postcode, "E-Class" — and it is unbeatable when you know the exact term. It fails the moment the listing says the same thing differently: search "estate car" and you will not find the one described as a "wagon".
Semantic search: flexible, and vague
Semantic search compares meaning by turning text into vectors and finding the nearest ones. "Estate car" and "wagon" land close together, so it finds what keyword search missed. But it has no notion of exactness: ask for a 2019 model and it will happily offer you 2018, because those are similar. That is the correct behaviour for meaning and the wrong behaviour for a filter.
Why merging beats picking
The two methods are wrong about different things, which is exactly the condition under which combining them helps. A listing found by both is very likely relevant. A listing found by only one is worth considering rather than discarding. The merge step keeps both kinds of evidence instead of throwing one away.
What hybrid search still does not fix
Merging improves recall — what gets considered. It does not by itself improve ordering. A hybrid system with no re-ranking still puts a plausible-but-wrong listing above a precise match, because nothing has yet checked the numbers. This is why serious systems treat retrieval and ranking as two stages, and why we do.
Where images come in
Text is not the only signal a listing carries. Photos encode things descriptions omit — light, condition, style, how a room is actually laid out. Treating images as a third retrieval alongside keyword and semantic is what makes "bright, with a lot of natural light" a searchable idea rather than a phrase someone has to have written down.