How do I stop my AI chatbot from making things up?

Ground it with RAG so answers come from your documents, instruct it explicitly to say when it does not know, keep responses cited and build an evaluation set of real questions you re-run on every change. Hallucination is managed with engineering, not hoped away.