A RAG chatbot is an AI chatbot that searches your own content, such as web pages, help articles and FAQs, before it writes each answer. RAG stands for retrieval-augmented generation: the chatbot first retrieves the passages that match the question, then the language model writes a reply based on those passages. This keeps answers specific to your business and up to date, without training a custom AI model.
| Plain AI chatbot | RAG chatbot | |
|---|---|---|
| Source of answers | What the model learned in training | Your own content, searched for each question |
| Knows your prices and policies | No | Yes |
| Stays current | Only when the model is retrained | As soon as your content changes |
| Risk of made-up answers | High for business questions | Lower; it can say it doesn't know |
| Shows sources | No | Can link to the page used |
Fine-tuning changes the AI model itself by training it on your examples. It is useful for tone or a fixed format, but it is slow and costly to update, and the model still can't quote a page it was never shown. RAG leaves the model as it is and gives it the right content for each question, which is why most website and support chatbots use RAG.
Newer chatbots combine RAG with tool calling: the AI decides whether to search your documents or call a system, such as a store's product catalogue. The Model Context Protocol (MCP) is a common way to connect these tools. See LooperChat's MCP integrations.
LooperChat is a RAG chatbot for websites. You add your URLs and FAQs, and it answers visitors from that content: