"I am only a human being and I have the right to be mistaken or misunderstood" - this sentence could be uttered by all of us, after all, we are not infallible. In the same way, chatbot may incorrectly answer the user's question. In order to efficiently capture such situations and verify whether the chatbot response was what the user expected, you can enable the Quality assurance mechanism (AI Consultant -> Quality assurance plugin).
After each automatic response, chatbot will ask the user whether the information he gave was useful. On the list of topics, you can watch on an ongoing basis for how many users the topic has been helpful, and for how many. If you notice that a topic has a lot of "uselessness" applications, verify the list of questions that point to the topic. Maybe your questions are too general or have you entered too short phrases?