AI Bots and hallucinations?

We are currently interested in implementing an Advance AI (EAP Generative AI specifically) bot in our help centres and before we I thought I'd ask if anyone else who has done so, has noticed any instances where their bot has delivered the wrong answer to the customers?  Are the bots having hallucinations?  How do you track the accuracy of the answers delivered to your customers?  Any advice based on your experience so far?