In one sentence
Retrieval-augmented generation (RAG) is an architecture in which a model searches a corpus for relevant documents and writes its answer from those documents. Most answer engines use some form of RAG, which is why what is published on the open web still decides what they say.
For a marketer the useful implication is narrow and concrete: retrieval happens at the passage level, not the page level. A model pulls a paragraph, not a document. Pages built as one long argument retrieve badly; pages built as a series of self-contained, clearly headed answers retrieve well.