Semantic search matches the meaning of a query rather than its exact words. Where keyword search finds assets whose metadata literally contains "celebration", semantic search also surfaces the photo tagged "team toasting at office party" — because the meanings are close, even though no word overlaps.

It works by converting both queries and asset descriptions into embeddings: numerical representations where similar meanings sit near each other. Searching becomes a proximity problem — find the assets whose vectors are closest to the query's — instead of a string-matching problem. The practical effect is that search stops depending on the searcher guessing the tagger's vocabulary. "Woman working from home" finds assets tagged "remote work", "laptop on couch", or captioned in another register entirely.

In a DAM, semantic search complements rather than replaces the other layers. Exact search still wins for product codes, filenames, and controlled terms; filters still handle structured constraints like format or license status. Semantic search covers the fuzzy middle — the conceptual queries that make up much of everyday asset hunting — and it gets better as enrichment adds more describable text to each asset: captions, transcripts, extracted document text.

In freedam, semantic search runs alongside full-text search in a hybrid ranking, using text and vision embeddings so natural-language queries match both what assets say and what they show.

Related terms: Visual Search, Metadata, OCR (Optical Character Recognition)