A plain AI model answers from what it learned in training, which is frozen in time and knows nothing about your company. RAG fixes that with a two-step move: first retrieve the relevant passages from a knowledge base (your documents, your help articles, your product data), then generate the answer using those passages as the source.
Whenever a tool promises "chat with your documents" or "an assistant trained on your knowledge base", the machinery underneath is usually RAG. The AI is not actually retrained on your material; it just gets handed the right pages at the right moment (into its context window).
Why you keep hearing it
Because RAG is how companies make AI useful with their own information, and because it reduces hallucination: an AI answering from a real document has less room to invent. Vendors say "RAG" constantly, roughly the way they used to say "cloud".
What it means for you
Two things. A RAG system is only as good as the documents behind it, so messy, outdated content produces confidently outdated answers. And your public content is increasingly retrieved by other people's AI systems too, which is one more reason clear, well-structured writing has become a distribution strategy.