A chatbot is a program that talks with you by text or voice. Some follow fixed rules, using menus or keywords that someone wrote in advance. Others use a language model (LLM) to interpret free-form sentences and write each answer on the spot.
A question from the app. Pick an answer.
A bot only gets you if you type "balance" or press 1, 2 or 3. What is behind it?
A language model would understand free-form sentences, not just "balance" or a number.
It only recognizes words or numbers someone planned for. Each one leads to a reply written in advance, and anything else gets the error reply.
Permissions limit what it does, not how it understands what you type.
The rule based bot is the "press 1 for your balance" kind. It follows a tree of options and keywords: type "balance" and it answers; type "how much money do I have left?" and, if nobody planned for it, you get "Sorry, I did not understand". It uses predefined replies, which can still be wrong.
A chatbot built on an LLM understands that sentence even if nobody planned for it, because it is not matching an exact word: it predicts an answer. The trade-off is that it can say something that sounds right and is false, a hallucination.
An LLM does not know your opening hours or your prices. A shop's bot usually looks them up in the business's documents before answering, with RAG. And the sensitive stuff, like a complaint or a cancellation, often still goes down a fixed path or gets handed to a person.
A chatbot is a conversational interface. It can explain how to cancel an order or cancel it if it has access. An AI agent can choose steps and tools to complete a goal.
The line is not that sharp: if your bank's bot really looks up your balance, it is already using a tool. The useful question is what it can do on its own, because that decides how much it needs checking.
"It learns from every chat." The model does not change while you talk: what it remembers from the chat is in its context. Whether the company later uses your chats to improve it depends on its policy. "Every chatbot uses an LLM." ELIZA, from 1966, replied using rules and keywords. "If it sounds sure, it is true." Sounding fluent does not make it right: check what matters.