Not sure the nature of the QA you perform but as somebody who was forced to use AI QA for a bit in a ticketing system for support purposes, my 2 cents is: don’t.

I did the exercise for 2 months where I double checked all the stuff it spewed as summaries of actions, recommendations, wrap up info and applied codes vs stuff that was actually happening in the tickets, it was consistently inconsistent and while on the surface as a quick check it looks fine, the moment any scrutiny is applied it falls apart.