Talking AI — RAG: forcing AI to consult your documents, not its memory

RAG (retrieval-augmented generation) means forcing AI to fetch the right documents from a repository — catalogs, past quotes, standards — before writing its answer, instead of drawing on its general memory. It is the standard technique for getting an assistant to cite real products and real prices rather than plausible approximations. The question to ask a vendor: "when the exact document doesn't exist in the repository, what does your system do — say so, or make it up?" The catch: RAG reduces errors, it does not eliminate them. Stanford tested the legal tools of the global leaders (LexisNexis, Thomson Reuters) sold as "hallucination-free" thanks to RAG: they still got 17% to 33% of answers wrong.

Source: Magesh et al., Stanford RegLab / HAI, "Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools" (Journal of Empirical Legal Studies, 2025).