What is Retrieval-augmented generation (RAG)?
Retrieval-augmented generation is a technique that lets a large language model pull in information from external documents before answering, so its reply draws on more than its training data.
Why does Retrieval-augmented generation (RAG) matter for marketers?
RAG is how many AI search and assistant products ground answers in web pages and company content. It explains why being retrievable, accurate and citable matters for brands, and why answers can still be wrong.
Where does Retrieval-augmented generation (RAG) stand today?
Wikipedia says the term was introduced in a 2020 paper by Lewis and others. Source data is converted to embeddings in a vector database, a retriever selects relevant documents, and these are added to the prompt. Benefits include using recent or domain-specific information without retraining and answers that can cite sources. Limits: RAG does not eliminate hallucinations, models can misread context, and Wikipedia reports a 2025 case of "RAG poisoning" in which a firm allegedly shaped online content for this purpose.
Common use cases
- Answering from company documents
- AI search with citations
- Grounding assistants in current information
Some systems add re-ranking, context selection or fine-tuning, and hybrid search combines vector search with traditional text search. Ars Technica is quoted on the page: “the LLM can still hallucinate around the source material.”
Why marketers meet it
Studies of AI answers, such as Semrush’s AI Visibility Index and Bing’s Topics Portfolio, measure which sites are cited, which is what a retrieval step selects. Different answer engines retrieve differently, which is why mentions and citations diverge.