The obvious solution is to implement a mock chatbot that answers from a set of pregenerated wrong answers. Noone will know the difference.