LOL. This matches my experience as a platform developer working with this tech for the past year for this same use case.
Extensive QA + red-teaming shows that Generative AI consistently leaks data from the training sets and fine-tuning layers, while also just plain making shit up.
None of my clients have deployed customer-facing generative chatbots for these reasons, instead opting to use older, much more tested conversational AI APIs for their bots. They 'stick to the script.'
Generative AI is pretty great at analyzing call transcriptions though, but extremely expensive due to the token cost.