It’s super useful, but you must be very careful about leading questions.
The AI will produce a set of output tokens that would logically follow given the input, so if you have deliberately or accidentally put a thumb on the scales in terms of the question asked, the output can often be directed in that direction.
It’s very similar to the issues pollsters have in designing their questions, a leading question however innocently crafted can bias the output of a pole in one direction or the other.
You can setup personas to strongman a particular take and then argue against that to test your points, but you need to be aware that at some point the number of tokens in the thread towards a particular viewpoint will start to affect the contents of the responses.