Many times, you will have a question that you think should be easily answered, but the model will return a non-answer. Consider the question: "Who fell from the loft?"
No matter what anyone tries to tell you, A.I. models aren't experts at anything: They don't know anything, they don't remember anything, they aren't human. What they are good at is retrieving information. Therefore, the more information, the more specifics, the more "details" you can give it about your question, the better the chances of your question being answered.
Granted, you wouldn't be asking the question if you knew the answer, you still should be able to provide as many specific details as possible. For instance, we know that the person who fell from the loft did so because he fell asleep. And we also know this happened during a sermon Paul was giving. Therefore, a more specific question would be:
"Who fell asleep and fell from the loft during the sermon?"
This, as you see, is easily answered by the model. If you think of A.I. as an expert, think of it as an expert in finding information, not knowing information. There is a difference.