Ask ten people how they use AI at work and eight will describe the same thing: type a question, read the answer, paste it somewhere. Then they are surprised when the result is bland or slightly wrong.
The other two describe something different. They treat the first answer as raw material. They push back on it, ask for the opposite view, request three versions and pick the best sentence from each. Their results are noticeably better, and it is not because they have better tools.
01The first output is the average
A model’s first answer is, roughly, the most probable answer. That makes it competent and unremarkable by design. Your job is to move it away from the average and toward the specific thing you need.
The first answer tells you what everyone would say. The third answer starts to sound like you.
02Three moves that work
Ask what is weakest. “What is the least convincing part of this?” gets a useful critique almost every time.
Ask for the opposite. If it recommended one approach, ask it to argue for the alternative, then decide yourself.
Give it your constraints. Word count, audience, what you have already tried. Specific inputs produce specific outputs.
03Where this breaks
This habit does not help when you cannot judge the output. If you are asking about a field you do not know, you will not spot the confident mistake in draft three any better than in draft one. Use the model to learn the field first, then to draft.
Pick one document you will write this week. Get a first draft from the model, then ask it three questions before you edit a single word. Notice how far the third version is from the first.