Jargon Buster / Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation (RAG).
AI & AutomationA method where AI first looks up your trusted documents, then writes its answer from what it finds.
What is retrieval-augmented generation?
Retrieval-augmented generation (RAG) is a technique in which an AI system first searches a trusted set of documents for relevant information, then uses what it finds to write its answer.
Rather than relying on what the model absorbed in training, RAG hands it the right pages from your own material at the moment of the question: your policies, price lists, product specifications or past proposals.
Why it matters
RAG is how a business gets AI that answers from its own knowledge without retraining a model. A staff assistant can answer HR questions from the handbook; a website assistant can quote your actual service areas and terms; a sales team can pull the right case details into a proposal. Because answers come from your documents, they can cite the source, which cuts hallucinations and makes checking quick. When a policy changes, you update the document and the answers follow.
How it works
- Prepare the content Your documents are gathered, cleaned and split into small sections.
- Index it Each section is converted into a searchable form so the system can find passages by meaning, as well as by keyword.
- Retrieve the passages When a question arrives, the system pulls the few passages most likely to contain the answer.
- Generate the answer The language model writes a reply using only those passages, ideally quoting or linking to them.
The honest view
RAG quality is mostly a content problem. If your documents are out of date, contradict each other or never wrote down the answer in the first place, the system will retrieve the wrong passage and answer it confidently. Many disappointing AI assistants have perfectly good technology sitting on top of a messy shared drive.
Retrieval can also simply miss. A question phrased differently from your documents may not find the right passage. The underrated work is editorial: one current version of each document, clear headings, and someone responsible for keeping them true.
What to do about it
List the documents your team answers questions from most often. Remove old versions, fix contradictions and give each document an owner and a review date. Then collect real questions from your inbox and test any RAG tool against them, checking both the answer and the passage it cited. If the citation is wrong, fix the content before blaming the model.
Related terms.
Hallucination
When an AI model states something false or invented with complete confidence.
AI & AutomationSee more LLarge Language Model (LLM)
An AI system trained on huge amounts of text that can write, summarise, answer questions and follow instructions.
AI & AutomationSee more CChatbot
Software that holds a text or voice conversation with people, usually to answer questions or route enquiries.
AI & AutomationSee more FFine-Tuning
Further training an existing AI model on your own examples so it behaves in a specific, consistent way.
AI & AutomationSee moreGot a question about your own setup? Let's talk.
Your first call is with our founder, Rhodri Williams.