A RAG chatbot combines two things: a retrieval system that searches your own content, and a large language model that turns retrieved passages into a readable answer. Instead of relying purely on what the model learned during training, it looks up relevant material first.
The typical pipeline has four stages: ingesting your documents, splitting them into chunks and converting them into embeddings, storing those embeddings in a vector database, and then, at query time, retrieving the most relevant chunks and passing them to the LLM as context.
The main benefit is grounding — answers can reference your actual policies, product docs, or internal knowledge instead of the model's general training data, which reduces (though doesn't eliminate) the risk of confidently wrong answers.
RAG works best when your knowledge base is reasonably well-structured and kept up to date, since retrieval quality directly affects answer quality.