Structured filtering is the application of conditions on structured, scalar fields — numbers, dates, categories, booleans — alongside vector similarity, so that search results satisfy both semantic relevance and concrete constraints. It is how vector search is married to the kind of precise filtering that traditional databases excel at.
Where vector similarity answers what is conceptually like this, structured filtering answers what meets these exact rules: published after a date, priced under a threshold, in a given category, owned by a particular user, marked as in stock. Real applications almost always need both, returning the most similar items that also satisfy hard business constraints.
The effectiveness of structured filtering depends on how the database evaluates the conditions during search — through pre-filtering, post-filtering, or in-graph filtering — and on supporting structures like payload indexes and bitmaps that keep filtered queries fast. Strong structured filtering, especially on selective or high-cardinality fields, is a defining capability of mature vector databases, since it is what lets semantic search operate within the precise boundaries that production use cases demand.