The claim
Content is retrieved by semantic meaning, so covering a topic's concepts matters more than exact-match phrasing.
What the evidence shows
Karpukhin et al. demonstrated that dense, embedding-based retrieval substantially outperforms traditional sparse keyword methods on open-domain question answering. Passages are matched by semantic similarity rather than literal term overlap. This underpins why comprehensive topical coverage, expressed naturally, retrieves better than keyword-stuffed text.