Let Claude write the prompt for you

Most people trying to use Claude (Anthropic's AI chatbot) spend ages writing the perfect prompt, then get frustrated when the answer misses the point. There is a simpler way: ask Claude to write the prompt for you.

This technique is often called reverse prompting. Instead of trying to guess what the AI needs to know, you let it question you first. Once it has the full picture, it drafts the instructions itself.

Why this matters

Writing a good prompt is hard. You have to think through edge cases, constraints and success criteria before you know whether the task will work. Reverse prompting turns that around. Claude asks the questions, you answer them, and the prompt writes itself.

How it works

Anthropic's best-practice guide for Claude Code (the version of Claude that builds and changes software) recommends that for larger pieces of work, you have Claude interview you first. Ask it to question you about how the work should be done, edge cases, concerns and trade-offs, and to dig into the hard parts you might not have considered.

Once the interview is finished, Claude writes a complete specification (a written description of what to build). You start a fresh session to carry out the work, so the new session is focused only on the task.

The guide says Claude can infer intent but cannot read your mind. The more precise your instructions, the fewer corrections you will need. Reverse prompting helps you give those instructions without having to think of everything yourself.

This works in the Claude chat, in Cowork (where Claude carries out a task using your files and apps) and in Claude Code.

How to do it

1. Tell Claude what you want to achieve, and ask it to interview you before it starts.

1. Answer its questions one at a time. Be specific where you can, and say plainly when you are unsure.

1. When Claude has enough information, ask it to write the full prompt it would need to do the job well.

1. Review the prompt. If anything is wrong or missing, correct it now.

1. Start a fresh session and paste the final prompt in to begin the work.

Here is a prompt you can copy:

> I want to [goal]. Before you start, ask me questions one at a time about anything you need to know. When you have enough, write the prompt you would want to receive to do this well, and show it to me before using it.

I used a version of this when guiding a software delivery partner on the prompt they were writing for Claude. Instead of drafting it from scratch, we had Claude question them and build the prompt through the conversation.

Common mistakes

Answering too quickly. If Claude asks about an edge case you have not thought through, take the time to work it out. The interview is where you catch problems, not after the work is done.

Skipping the fresh session. If you run the final prompt in the same chat as the interview, Claude still has all the earlier discussion in its memory. That can help, but it can also mean the work drifts from the specification. A clean start keeps it focused.

Not reviewing the prompt Claude writes. It will be good in most cases, but it is your job to check it matches what you meant.

One thing to try this week

Pick a task you were going to ask Claude to do. Before you write the prompt yourself, ask Claude to interview you instead. See whether the prompt it writes is better than the one you had in mind.

If you want to see what else your business could automate, take the free What could your business automate? check.