Glossary · AI and automation
What is retrieval-augmented generation (RAG)?
Retrieval-augmented generation (RAG) is a technique in which an AI system first searches a trusted knowledge source for passages relevant to a question, then gives those passages to a language model to write the answer. In customer support, RAG keeps replies tied to the business's current policies and help content instead of the model's general training data.
Also called: RAG, retrieval augmented generation
On this page
Retrieval-augmented generation (RAG), explained
A language model on its own answers from patterns learned during training. It does not know your returns window, your shipping cut-off or last week's price change. RAG solves this by adding a retrieval step: the system turns the customer's question into a search, pulls the most relevant snippets from your help articles, policies or product pages, and instructs the model to answer using those snippets.
The approach was named in a 2020 research paper by Lewis and colleagues at Facebook AI Research, which combined a neural retriever with a text generator for knowledge-intensive tasks. It has since become the standard pattern for support bots because it lets a business update answers by editing content rather than retraining a model.
RAG quality depends mostly on the retrieval half. If the right passage is missing, outdated or split awkwardly across chunks, the model will either answer from partial information or guess. Good systems show which source they used, so customers and reviewers can check it, and fall back to a human when nothing relevant is found.
RAG reduces hallucination but does not eliminate it; models can still misread or over-generalise a correct source.
How to measure it
Test retrieval and generation separately. For a set of real questions, check whether the correct source was retrieved (retrieval hit rate) and whether the answer is supported by that source (grounded answer rate).
Common mistake
How SuperBot relates
SuperBot answers through layered retrieval: your training Q&A first, then knowledge sources (URLs, pasted text, documents), then published help articles, then a constrained model reply, and human handoff when confidence is low. Replies show a source label when matched knowledge supports the answer.
Frequently asked questions
What is RAG in customer support?
It is the pattern where a support bot searches your help content for the passages that match a customer's question and has the language model answer from those passages, so replies reflect your current policies.
Does RAG stop AI hallucinations?
It reduces them by giving the model the right facts, but it does not stop them completely. The model can still misread a source, and retrieval can miss. Show sources and keep a human fallback.
Do I need to fine-tune a model to use RAG?
No. RAG works with a general-purpose model; you change answers by editing the knowledge it retrieves from, which is faster and easier to audit than retraining.
Sources
Test it on your own questions.
SuperBot trains on your content, answers with citations and hands off to your team. The free plan includes 200 conversations a month, no card required.