Every time you look at something AI produced and think "that's not what I meant," the instinct is to learn better prompting. Find a framework. Add more detail.
But the problem is not the prompt. It is upstream. What you meant was never in words. It lived in your experience, in years of pattern recognition you stopped being conscious of a long time ago.
You fix it by learning to articulate what you already know. Three questions do the job.
The broken robot test
"If a very smart but completely literal person did this task, what would they get wrong?"
This question names your invisible rules. The ones you follow automatically. The ones you would never think to mention because they are obvious to you and completely non-obvious to everyone else.
Think about a time when you delegated something and got back work that was technically correct but clearly wrong. The person followed your instructions exactly. They just missed the things you did not say. That gap is what the broken robot test fills.
The trade-off test
"If I could only have one (quality A or quality B), which would I pick?"
This reveals your actual value hierarchy. Everyone says they want output that is thorough AND concise. Detailed AND fast. In reality, you always have a preference. When forced to choose, you pick one side. That preference is invisible knowledge that shapes every evaluation you make.
The success picture
"Forget the steps. What does the world look like when this is done perfectly?"
Most delegation instructions describe activities. Do this, then this, then this. But activities without a clear picture of success produce output that checks every box and misses the point. The success picture forces you to describe the real outcome, not the task.
Watch what the questions reveal
Say you are delegating a client meeting summary. Without the questions, you write: "Summarize the key points. Keep it to one page."
After the broken robot test: "A literal person would write a balanced recap of everything discussed. But I need emphasis on commitments made and open risks. If nobody committed to anything, I need to know that explicitly."
After the trade-off test: "If I had to choose between comprehensive and actionable, I pick actionable. I would rather have a short summary that tells me what happens next than a thorough one I have to re-read."
After the success picture: "When this is done right, I can forward it to the client within 10 minutes, and both sides can look at it a month from now and know exactly what was agreed and who owns what."
Read both versions. Same person. Same knowledge. Four minutes of making the invisible visible.
The bottom line
Stop chasing better prompts. Start chasing better self-awareness. The three questions take less than five minutes and surface knowledge you have been carrying unconsciously for years. Try the broken robot test on one task this week. Put the answer in your instructions. See what changes.