Its because how AI works, You build a perfect prompt with all the details, but you are basically giving a free will to LLM, so it might consider all the point or might skip some.
Whereas, in questionaries its a statement approved or rejected by a human. Hold a greater value then a vague detailed prompt. Here each statement has clear outcome defiened and discussed. No scope for hallucination. You correct the AI in first place even if it misses something or misunderstood thats the whole purpose of questionaries to bring the Clarity not just to AI. At the end it becomes the only ground truth for both.
Leaving no scope for LLM to wonder. If it needs to wonder it will leave a room for discussion, like a real engineer would do.