What is a chatbot and how to build one?

A useful one is mostly retrieval, not generation. The user's question is used to find the relevant passages in your own content, those passages are passed to the language model along with an instruction to answer only from them, and the answer is returned with its sources. The model supplies the phrasing; your data supplies the facts. Skipping the retrieval step is what produces a bot that answers confidently about things your business does not do.

The two things teams underestimate are evaluation and escalation. Without a fixed set of real questions and expected answers, you cannot tell whether a prompt change made it better or worse — you are guessing from anecdotes. And without a clear route to a human, every question outside its knowledge becomes a dead end, which costs more goodwill than not having a bot at all.