What is RAG?
RAG (retrieval-augmented generation) means the model looks up your documents first, then writes the answer from what it found. It is a search step plus a writing step.
Also called: retrieval-augmented generation
In plain English
Without RAG, the model answers from memory and can sound sure while being wrong. With RAG, you point it at a folder: your price sheet, your SOPs, last month's report. It pulls the relevant pages, then answers. If the folder is messy, the answer is messy. Name the folder out loud before you turn retrieval on. A price sheet from March and a price sheet from last week will both come back, and the model will not know which cashier is using. Put a date in the file name. If two files disagree, the page should say which one wins.
How an owner would use it
Put the files you trust in one place and tell the tool to use only those. Ask it to quote the page it used. If it cannot point at a page, do not send the answer to a client.
Common mistakes
People bolt RAG onto a messy shared drive and then blame the model. Retrieval only finds what you stored. Old fee sheets, three versions of the same SOP, and a folder of customer files will all come back. Labels like graph RAG or agentic RAG are still this job: pick a source, pull a passage, then write. If you will not clean the folder, do not turn retrieval on for customer replies.
A worked example
A small marketing agency
Answers from the client folder
This is an illustration, not a customer result. A producer asks, 'What did we promise this client on revisions?' The agent searches that client's folder, quotes the proposal line, and drafts a reply. It does not answer from a general idea of how agencies work.
Related terms
Prompts and guides
Questions owners ask
What is RAG?
RAG means the AI retrieves your documents and then generates an answer from them. The lookup is the part that keeps the answer tied to your business.
How do I use RAG in a small business?
Collect the files that are actually true (current prices, SOPs, proposals). Point the assistant at that set. Require a citation. Update the file when the rule changes, or the assistant will quote the old one.
When is RAG the wrong fix?
If the answer is a single standing rule, put it in the instructions instead of a document search. If the failure is tone, retrieval will not teach voice. Retrieval helps when the facts live in files that change.
Product guides are on AI apps for business.