The claim
AI answer engines depend on a retrieval step that selects source documents, so being retrievable and quotable is a precondition for being cited.
What the evidence shows
The foundational RAG paper by Lewis et al. demonstrates that combining a parametric language model with a non-parametric retrieval index improves performance on knowledge-intensive tasks and reduces hallucination. This establishes that answer quality - and which sources get surfaced - hinges on the retrieval layer selecting relevant, well-structured passages.