Prompt | Prompting

Stop writing prompts. Make the AI write them.

One line after your rough prompt. The AI asks you 3 or 4 questions, then hands you the finished version. You paste it and run it. No course, no framework to memorize.

Start here

Why you can skip the prompting course

You've heard you need to learn prompt engineering. It sounds like a course, and you don't have time for a course.

You don't need one. Write the prompt the way you'd say it out loud, rough and short. Then ask the AI to rewrite it, with one condition: it has to ask you questions first.

It comes back with 3 or 4 questions. Who is this for. How long should it be. What should it never do. Short answers are fine. Then it hands you a finished prompt with everything you forgot to say already in it.

ChatGPTClaudeGemini1 lineNo setup
The line

Paste this after your rough prompt

Type your rough prompt first. Hit return twice. Then paste this line and send.

Copy it.
Rewrite this prompt to make it better. Ask me clarifying questions first, then give me the improved prompt.

Or copy the whole thing in one go and fill in the bracket.

Copy it. Fill in anything in [brackets].
[Your rough prompt, written the way you'd say it out loud]

Rewrite this prompt to make it better. Ask me clarifying questions first, then give me the improved prompt.
Before you paste it: don't polish the rough version

That's the AI's job. If you spend 10 minutes on the first draft, you've done the work the line was supposed to do for you. One sentence is enough. Typos are fine.

Before and after

The same request, both sides

Say you keep the books for a handful of small businesses and your rate goes up in January. This is the prompt most people would type.

Before: what you'd type

Write an email to my clients telling them my prices are going up.

You add the line and send. This is what comes back, with the kind of short answers that are enough.

What it asks, and what you answer
Q1Who are these clients, and how long have most of them been with you?
YouSmall business owners on monthly bookkeeping. Most have been with me 2 to 5 years.
Q2How much is the increase, and when does it start?
You$350 to $400 a month, starting January 1.
Q3What tone do you want, and how long should it be?
YouWarm and direct. Under 150 words. Sounds like me, not a company.
Q4Is there anything the email should never do?
YouNo apologizing. No long explanation of why. No discount offer.
After: the prompt it hands you
Copy it.
Write an email from me to my monthly bookkeeping clients, small business owners who have been with me 2 to 5 years, letting them know my monthly rate is going from $350 to $400 starting January 1.

Tone: warm and direct, the way I'd say it in person. Under 150 words.

Structure: open with a thank-you for their trust, state the new rate and the date in one clear sentence, say what stays the same (their bookkeeper, their turnaround, their monthly call), and close by inviting them to reply with any questions.

Do not apologize for the increase. Do not explain why beyond one short sentence. Do not offer a discount or an exception. No exclamation marks. Sign off with my first name only.

Look at what it added that you never said: who the reader is, the exact numbers, a structure, what stays the same, and 4 rules. That's the gap between a generic email and one you'd send. You typed 1 sentence and 4 short answers.

The routine

Rough, rewrite, run

3 steps, the same every time. After a week it takes less time than writing the prompt yourself did.

Rough
Type the request the way you'd say it out loud. One sentence. Don't fix it.
Rewrite
Paste the line and send. Answer its questions with short answers. If a question doesn't matter to you, say "assume the rest" and move on.
Run
Copy the rewritten prompt and paste it into a fresh chat, then send. A fresh chat keeps it clean, since your answers are already folded into the prompt. The same chat works too.
What happens after a few weeks

You start to see what it adds every time. The audience, the format, the limits, what to leave out. And you start writing that way on your own, without trying. The AI taught you prompting by fixing yours.

Keep the good ones

Save the rewritten prompts you'll use again

The rewritten prompt is the reusable part. The price increase email happens once a year. The client update, the invoice nudge, the caption for a video, those come back every week. Don't rebuild them every time.

When it goes wrong

4 things that happen, and what to type

It rewrote the prompt without asking anything

Type: "Don't rewrite yet. Ask me the questions first." Some models skip the questions when the prompt looks complete. Make it ask anyway. The questions are where the missing details show up.

It asked 8 questions

Answer the 3 that matter and type "use your best judgment for the rest." You don't owe it an essay. It fills the gaps with sensible defaults, and you fix anything you don't like in the result.

The rewritten prompt is long and sounds formal

That's fine. It's written for the AI to read. If the result comes out stiff, the fix goes in the prompt: add "write like a person, in plain words" to it and run it again.

The result is still off

Fix the prompt, then rerun it. Tell it what went wrong (too long, it apologized, it added a discount) and ask for the prompt again with that rule added. Save that version. Next time it's right on the first run.

Rather tick boxes?

The click-instead-of-type option

If answering questions by typing is the part you dread, there's a free site built around this exact move: Metaprompt.com. You paste your rough prompt, it turns its questions into checkboxes and dropdowns, you tick what fits, and it hands you the rewritten prompt. Copy it into ChatGPT, Claude or Gemini, or run it on the page.

2 things to know before you go. It asks you to make a free account (with YouSpot, the company behind it) before you can use the tool. And you don't need it. The line above does the same job inside the AI you already use. The site is for people who'd rather click than type.

Metaprompt.com
Free, needs a free account
Open it →