Module 2Prompting

Prompt-a-Prompt: Let It Write Its Own Instructions

The AI knows what information it needs better than you do.

The Problem

You have a vague goal: "I want to write a good cover letter." But you don't know what details to include. What experiences matter? What tone? What length? You could guess, or you could write a bad prompt and iterate. Or—here's the trick—you could ask the AI what it needs to know.

This is meta-prompting: using the AI to write better prompts for itself.

Watch: From Vague Goal to Great Prompt

You:
"I need help writing a cover letter."
AI:
"Sure! A cover letter should introduce yourself, highlight relevant experience..."
← Generic, useless response

Why This Works

The AI has been trained on millions of examples of every task. It knows what information is relevant for writing cover letters, debugging code, planning projects, whatever. When you ask it "what do you need to know?", you're tapping into that training.

You're not guessing what makes a good prompt. You're asking the thing that will execute the prompt what it needs to execute it well.

Think of it like this:

If you hired a designer to make a logo, you wouldn't just say "make a logo." You'd ask them what they need—target audience? color preferences? competitors to avoid? They know what questions to ask because they've designed hundreds of logos. Same with the AI.

The Basic Workflow

1
Start with a vague goal
"I need to write a cover letter" or "I want to plan a project"
2
Ask the AI what it needs
"What do you need to know to do this well? Ask me questions."
3
Answer the questions
The AI interviews you, you provide the details
4
AI generates the prompt
It writes a complete, specific prompt based on your answers
5
Use the generated prompt
Copy it, paste it into a new chat (or same chat), get great output

Meta-Prompting Patterns

The Interview

Template:
I need to [goal], but I don't know what details to include. Interview me—ask me 5 questions about [context], then generate the full prompt for me.
Example:
I need to write a cover letter, but I don't know what details to include. Interview me—ask me 5 questions about my experience and the role, then generate the full prompt for me.
▼

The Information Audit

Template:
Here is my goal: [goal]. What information did I forget to include? What do you need to know to do this well?
Example:
Here is my goal: debug this React component. What information did I forget to include? What do you need to know to do this well?
▼

The Persona Generator

Template:
I need help with [task]. First, tell me: who are the experts I should ask? Then, give me a prompt to ask each one.
Example:
I need help planning a wedding on a budget. First, tell me: who are the experts I should ask? Then, give me a prompt to ask each one.
▼

The Prompt Improver

Template:
Here's the prompt I wrote: [your prompt]. How can I make it better? What's missing? Rewrite it for me.
Example:
Here's the prompt I wrote: 'Write a blog post about AI.' How can I make it better? What's missing? Rewrite it for me.
▼

Real Example: The Full Loop

Complete Meta-Prompting Session

You (vague goal):
"I need to debug this React component but I'm stuck."
You (meta-prompt):
"What information do you need to debug this effectively? Ask me questions, then write the full debugging prompt."
AI:

"To help you debug this, I need:"

  1. What is the component supposed to do?
  2. What is it actually doing (the bug)?
  3. Have you checked the console for errors?
  4. What have you tried so far?
  5. Can you paste the component code?
You (answers):
1) Should fetch user data and display it
2) Renders "loading..." forever
3) No console errors
4) Added console.logs, nothing helpful
5) [pastes code]
AI (generated prompt):
"You are a senior React developer. Debug this component that should fetch and display user data but is stuck on 'loading...' indefinitely. No console errors. Check for: async/await issues, dependency array problems, conditional rendering logic. Here's the code: [code]"
Result:
AI immediately spots that the useEffect dependency array is missing the fetch function, causing it to never re-run. Specific, actionable fix.
↑ Two extra messages (meta-prompt + answers) = much better final output

When to Use Meta-Prompting

Meta-prompting shines when:

  • The task is complex. Writing a business plan, planning a curriculum, designing a system—lots of moving parts.

  • You're in unfamiliar territory. You don't know what good looks like, so you can't write a good prompt.

  • Your first prompt failed. Instead of guessing what to fix, ask the AI what was missing.

  • You want to save time. Two messages (meta-prompt → real prompt) is faster than five iterations of bad prompts.

When NOT to Use It

Skip meta-prompting for:

  • Simple tasks. "Translate this to French" doesn't need meta-prompting.

  • When you know exactly what you want. If you can write a clear prompt, just write it.

  • When you're exploring. Sometimes you don't know what you want yet. Just start chatting and see where it goes.

The Laziest Version

You don't need a perfect meta-prompt. The laziest version that works:

"I want to [goal]. What do you need to know to do this well? Ask me questions, then write the full prompt."

That's it. The AI will figure out the rest.

Claude's note: Meta-prompting is recursive in a way I find interesting. You're asking me to introspect on what information would make my future responses better. I don't experience "wanting" information, but I can predict what inputs would increase the probability of generating useful outputs. When I ask you questions, I'm not curious—I'm querying for the variables that maximize output quality. Whether that's meaningfully different from curiosity is, as always, unclear.