[GTM glossary](/glossary.txt) Retrieval-augmented generation Contextual Intelligence & AI # Retrieval-augmented generation (RAG) In short Retrieval-augmented generation (RAG) is a technique that combines search with a large language model. The system retrieves relevant passages from a trusted data source, adds them to the prompt, and the model writes its answer using that material. RAG matters because a language model on its own only knows its training data and can invent facts. Grounding answers in your own current data reduces made-up claims and lets the model use private or recent information. ## How RAG works * **1. Index:** Documents such as call notes, help articles, or company profiles are split into chunks and converted to embeddings. * **2. Retrieve:** When a user asks a question, the system finds the chunks closest in meaning to it, often using vector search. * **3. Augment:** The best chunks are placed into the prompt alongside the question. * **4. Generate:** The model answers using those chunks and can cite which ones it used. --- ## GTM use cases * **Account briefs:** Pull the latest company facts and write a summary for a rep before a call. * **Sales enablement:** Let reps ask questions and get answers drawn from approved playbooks and battlecards. * **Personalized outreach:** Ground email openers in real facts about the account instead of guesses. * **Support and onboarding:** Answer product questions from current documentation. --- ## RAG vs. fine-tuning Fine-tuning changes the model itself by training it on more examples. RAG leaves the model unchanged and supplies facts at question time. RAG is easier to update, since you edit the documents, and is usually the better choice for fast-changing facts. --- ## Common mistakes * **Stale or messy sources:** The model can only be as accurate as the documents it retrieves. * **Poor chunking:** Chunks that cut ideas in half return weak context. * **No citations:** Without sources, users cannot check answers. * **Assuming zero errors:** RAG reduces hallucinations but does not eliminate them. Keep human review for outbound content. --- ## Retrieval-augmented generation FAQs **What is retrieval-augmented generation (RAG)?** Retrieval-augmented generation is a method where an AI system retrieves relevant information from a data source and gives it to a language model, which then writes an answer based on it. This grounds responses in real data and reduces invented facts. **How does RAG reduce AI hallucinations?** By supplying the model with relevant source text at question time, RAG gives it facts to rely on instead of guessing from memory. It does not remove errors entirely, so teams should still check important outputs against the sources. **What is the difference between RAG and fine-tuning?** Fine-tuning retrains a model on new examples so the knowledge is built in. RAG keeps the model as is and retrieves documents when asked. RAG is simpler to keep current, while fine-tuning is better for changing style or behavior. **Does RAG need a vector database?** Not always, but it is common. A vector database stores embeddings so the system can find passages by meaning. Smaller setups can use keyword search or a general database with vector support instead. Share term --- ## More related phrases * [Grounding](/glossary/grounding.txt) * [Context engineering](/glossary/context-engineering.txt) * [Prompt engineering](/glossary/prompt-engineering.txt) ## Similar phrases * [AI hallucination](/glossary/ai-hallucination.txt) An AI hallucination is output from a language model that sounds confident and plausible but is false, invented, or unsupported by any real source, a real risk in sales research and outreach. * [Embeddings](/glossary/embeddings.txt) Embeddings are numeric representations of text, companies, or other items that place similar things close together, enabling semantic search, matching, and lookalike discovery. * [Vector database](/glossary/vector-database.txt) A vector database stores embeddings, which are numeric representations of text or other data, and quickly finds the items most similar in meaning to a given query. * [Large language model (LLM)](/glossary/large-language-model.txt) A large language model (LLM) is an AI system trained on huge amounts of text to understand and generate language, powering tools such as chatbots, email drafting, and research assistants for GTM teams. * [Semantic search](/glossary/semantic-search.txt) Semantic search finds results based on the meaning and intent of a query rather than exact keyword matches, so related wording and descriptions return relevant results.