Prompt Basics · Lesson 2 of 3
How to refine
The first answer is almost never the right answer.
Refining is the skill. Beginners treat the first reply as the answer. Skilled users treat it as a draft and reshape it in two or three small turns. Three moves cover most of what you need.
The first answer AI gives is shaped by averages. It writes the kind of reply most people seem to want, not the one you actually want. The way to fix that is not a longer first prompt. It is a short second message that reshapes what you got.
Think of it like working with a junior who is fast but generic. You do not rewrite their draft from scratch. You tell them what to change. That is what refining is.
The three moves
- 1
Make it shorter
AI writes long because long looks thorough. It is usually noise. Cutting forces the model to pick what matters. The shorter version is almost always better.
When to use:
When the answer feels heavy or padded.
Try saying:
Make it half as long. Keep only the parts that actually help.
- 2
Make it more specific
Generic advice is generic because the prompt was generic. Push back. Paste your actual situation. The AI cannot read your mind, but it can apply patterns to facts you give it.
When to use:
When the answer sounds like a textbook, not advice.
Try saying:
Now apply that to my actual situation: I am [your role], the problem is [your problem], the constraints are [your constraints]. What should I do?
- 3
Show me three versions
When you cannot tell if the AI is right, asking for three options exposes the range. Often the second option is the one you actually wanted. You learn what you prefer by comparison.
When to use:
When you are stuck between trusting the answer and rewriting it.
Try saying:
Give me three different versions. One short and direct. One warm and detailed. One that takes a different angle. Number them.
A fourth move, for later
“What would change your mind?”
When AI sounds confident but you suspect it is wrong, ask it what would change its mind. Or what it is assuming. This forces the model to surface the hidden assumptions inside its answer. Often that is where the mistake lives.
Try saying:
What are you assuming when you give that answer? What would have to be true for that answer to be wrong?
Worked example — three refines
Prepping for a job interview at an Indian multinational where the panel includes overseas colleagues.
First prompt
Help me prepare for a job interview.
First answer: Ten generic questions about strengths, weaknesses, “tell me about yourself”. Useless because every interview asks these.
Refine 1 — more specific
Apply that to my actual interview: senior product manager role at an Indian fintech, panel includes overseas colleagues, focus on payments. What questions are likely?
Now: Questions about UPI volumes, fraud trade-offs, working across time zones. Specific. Useful.
Refine 2 — shorter
Cut to the five most likely questions. One line each.
Now: Five questions you can actually prep answers to before the interview.
Refine 3 — three versions
For question 1, give me three different ways to answer. One short. One with a story. One that flips the question back. I will pick.
Now: Three real options. You pick the one that fits your style. The interview prep is done in 15 minutes.
Try it yourself
Open ChatGPT or Claude. Ask a generic question. Then refine it three times using the three moves. Notice the difference between the first answer and the third refine.