Most people type a prompt into ChatGPT or Claude, cross their fingers, and take whatever comes back. When it's underwhelming, they quietly decide AI is "a bit overhyped." There's a 30-second habit that changes all of it — and it's the opposite of what you'd expect.
You don't write a better prompt. You ask the AI to write it for you.
It's called meta prompting, and I picked it up from Dharmesh Shah — the founder of HubSpot, who writes the excellent simple.ai newsletter. It was one of those "why did nobody tell me this sooner" moments. He rates it so highly he built a free tool around it: Metaprompt.com.
The trick, in one sentence
Instead of agonising over the perfect prompt, paste your rough one and ask the AI to rewrite it — and to ask you a few clarifying questions first.
That's it. You're not writing the prompt anymore. You're handing the AI your messy first attempt and letting it do the part that actually takes skill.
Why this is perfect if you're "not technical"
The craft of writing good prompts has a slightly intimidating name — prompt engineering — and a whole world of frameworks and acronyms behind it. Most people take one look and decide it's not for them.
Meta prompting skips all of that. You don't learn the frameworks. You let the AI apply them to your words. It's the difference between studying grammar for a year and just asking a brilliant editor to tighten your draft.
The magic is in the questions
The real unlock is giving the AI permission to ask you things first. Who's the audience? How long? What tone? What's the one thing you want the reader to do?
You answer a couple of quick questions — and suddenly the model has everything it needs. That little back-and-forth is the difference between a bland answer and one that sounds like it came from you.
You're not learning to prompt better. You're letting the AI interview you — and that's what makes the result good.
It's also the fastest way to get better at this
Here's the bonus. Every time you watch the AI rewrite your scrappy prompt into a good one, you absorb a little of how it's done — the value of context, of specificity, of saying what you don't want.
After a few weeks of this, you'll notice your first drafts getting better on their own. You'll have learned prompt engineering without ever sitting down to "learn prompt engineering."
One model, learned well
One thing worth saying: pick a capable model and actually get to know it, rather than hopping between five. Dharmesh starts most of his work with GPT-5. I've built this whole website, my AI tools, and the little chatbot in the corner of this page with Claude — it's exceptional at following instructions, which is exactly what meta prompting leans on.
The point isn't which one. It's that this trick works on whatever you've already got open. Pick one, get comfortable, and let it fix your prompts.
Try it on your very next prompt
You don't need a special tool to start (though Dharmesh's Metaprompt.com makes it even easier — it turns the clarifying questions into simple checkboxes). Just paste this in front of any request:
"Before you answer, rewrite the prompt below so it's clearer, more specific, and more likely to get a great result — and ask me up to three clarifying questions first. Here's my prompt: [paste your rough prompt]"
Do that once and you'll feel the difference immediately. Do it for a fortnight and you'll never go back to typing-and-hoping.
The bigger point
Getting good with AI isn't about memorising clever tricks. It's about a few small habits that compound — and being willing to let the tool help you use the tool. That's exactly the kind of thing we work through together in Build Made Simple: practical, hands-on, no jargon, for people who'd never call themselves "techy."
Credit where it's due: the meta-prompting technique and the Metaprompt tool are from Dharmesh Shah over at simple.ai. I just couldn't keep this one to myself.
