How it works in a data room
Documents are converted into numerical representations of meaning, often called embeddings, alongside the usual keyword index. A query is converted the same way, and the system returns passages whose meaning is closest, even when they share no words with the query. Many rooms combine semantic and keyword results so that exact phrases, names and numbers still surface reliably. Scanned files need OCR first, and results must respect each user’s permissions.
Why it matters in a deal
Contracts and reports use inconsistent language: one lease says “break option”, another “early termination right”, a third “tenant may end the lease”. Keyword-only full-text search misses some of them unless the reviewer guesses every variation. Semantic search narrows that gap and is the retrieval layer behind most AI assistants. It can also return loosely related results, so exact-match search remains useful for defined terms and figures.
Example
A buyer reviewing 2,000 customer contracts searches for “most favored customer pricing”. Keyword search finds 9 contracts. Semantic search finds 23, including clauses phrased as “prices no less favorable than those offered to any other buyer”. Counsel reviews all 23 and flags four as material.