Works okay until policies or other variables change. Then it's just confidently providing bad information which results in additional support time. And when it'

Works okay until policies or other variables change. Then it's just confidently providing bad information which results in additional support time. And when it's wrong/lying 10% of the time what's the liability impact - what if the question is about handling allergens in food? Also updating that kind of knowledge base is going to be a nightmare.