Meta Prompting: The Secret to Better AI Results

I've been experimenting with AI prompting techniques for years now (time flies!), and there's one useful technique that I'm genuinely surprised isn't more widely known.

It's called "meta prompting," and it's simple: you ask the AI to rewrite your prompt and make it better, giving it permission to ask clarifying questions that will make the prompt more specific.

This technique works so well that I actually built a tool around it: Metaprompt.com.

But before we get into the tool, I want to break down:

• Why choosing the right model matters before you start prompting

• How meta prompting actually works (with examples)

• Why this technique deserves a spot in your AI workflow

—@dharmesh

Choosing The Right Model

Back in 2022, when ChatGPT launched, picking a model was easy -- there was basically one option that worked well (ChatGPT).

Fast forward to today, and we have serious contenders like Anthropic's Claude, Google's Gemini, plus a growing number of capable open-source models. The landscape has become much more complex.

The good news is that things got simpler recently with the release of OpenAI's GPT-5. It has become my go-to model for most things, and that's what I recommend unless you have a special reason to use another model.

Here's why GPT-5 works so well for majority of my use cases:

• It follows instructions exceptionally well (crucial for meta prompting)

• It supports deep reasoning for complex problems

• It has great tool support for integrations

• It includes native connectors for platforms like HubSpot :)

That said, I still switch models based on specific needs:

• For coding tasks, I still occasionally use Claude

• For creative writing, Claude sometimes provides more interesting, human-like perspectives

• For analysis of large documents, Gemini’s longer context window can be valuable

But for 80%+ of my AI interactions, GPT-5 delivers the most consistent, high-quality results. Start there, then experiment with others as you get more comfortable with the basics.

How Meta Prompting Works (And Why It's Magic)

Now that we have a model picked, here's the pro tip technique that works really well at crafting perfect prompts: meta prompting.

The concept is simple but highly effective. Instead of struggling to write the perfect prompt yourself, you ask the AI to rewrite your prompt and make it better. You give it permission to ask any clarifying questions that will make the prompt more specific.

You can do this manually with any system, but I found myself using this technique so often that I decided to build a dedicated tool for it: Metaprompt.com.

Here’s how the tool works:

Step 1: Write your initial prompt (even if it's rough or incomplete)

Step 2: Normally, you prompt the system to ask you questions about your prompt to make it more effective. Then it generates a series of questions for you to answer first. But if you use Metaprompt.com, the tool generates optimization questions and presents them as simple checkboxes - no need to type lengthy responses.

Step 3: Based on your selections, it creates an optimized prompt automatically, and you can either copy the optimized prompt, or run it with GPT-5 directly on page.

The difference in output quality when you use the tool is dramatic. A rough initial prompt gets you generic advice, while the optimized prompt gives the model far more context, getting you better responses with the right tone and specificity.

What I love about turning this into a tool is that it makes meta prompting more accessible to anyone -- even to those just trying ChatGPT for the first time. No need to study prompting techniques or memorize frameworks, just click a few checkboxes and get better results immediately.

Why This Technique Deserves a Spot in Your AI Toolkit

The whole craft of writing good prompts is known as prompt engineering, and it's been around for a while now. But most people find prompt engineering intimidating or overwhelming.

Meta prompting solves that problem. It's a way for anyone -- even someone trying ChatGPT for the first time -- to immediately start getting better results without studying prompting techniques or memorizing frameworks.

If you're a power user like me who interacts with AI multiple times per day, you won't always need to meta prompt because you develop an "AI intuition" for what works. But I do find myself using this technique often, especially when tackling something outside my usual use cases.

What's particularly valuable is how meta prompting functions as a learning tool. By watching how AI systems restructure and improve your prompts, you naturally absorb what makes prompts more effective.

You start noticing patterns: the importance of context, the power of specificity, the value of clear constraints and desired outcomes (more on all this later!).

Over time, you'll find yourself writing better initial prompts because you've internalized these lessons.

Happy prompting!

—Dharmesh (@dharmesh)