The claim
Retrieval systems match at the passage level, so content structured into coherent, self-contained passages is more retrievable than sprawling undifferentiated text.
What the evidence shows
Dense Passage Retrieval (Karpukhin et al.) shows that dense vector retrieval over passages outperforms sparse methods for open-domain QA, and that passages are the retrieval unit. Passages that state their context and answer a discrete question without relying on distant surrounding text are more likely to be matched and returned.