Study Guide · Chapter 4: RAG Grounding · 5 min read

RAG, Explained Simply

Ask a closed model about your company's vacation policy and it will happily invent one. Retrieval-Augmented Generation (RAG) fixes this the same way an open-book exam fixes guessing: let the model look up the real answer before responding.

The open-book exam analogy

A closed-book exam forces a student to rely on memory — and memory can be wrong or outdated. An open-book exam lets them find the actual passage, read it, and answer based on what it says.

RAG does exactly that for models. Your question is matched against a searchable library of trusted documents, the best passages are pulled out, and the model answers using those passages as its source material.

Why grounding beats memorization

Model weights freeze at training time. Everything that happened after — new prices, new policies, new research — simply is not in there. Retrieval injects that freshness on demand.

Grounded answers also carry receipts. Because the response comes from retrieved passages, the system can cite sources, letting users verify claims instead of trusting a black box.

Where RAG fits

Anytime accuracy matters more than creativity — policy questions, contract analysis, technical support, internal search — RAG is the default architecture.

It complements rather than replaces good models: the model supplies language skill and reasoning; retrieval supplies current, specific, trustworthy facts.

Key Points

  • RAG = retrieve relevant documents first, then generate an answer grounded in them.
  • Retrieval adds knowledge the model never learned and keeps it perpetually fresh.
  • Citations turn 'trust me' answers into verifiable ones.
  • Use RAG whenever being right matters more than being creative.

All study guides for this chapter: RAG, Explained Simply · How RAG Works Under the Hood · RAG in the Real World