Stop Writing Prompts. Have ChatGPT Do It Instead.

How much time did you spend last week trying to craft the perfect prompt?

If you’re like most marketers I talk to, it’s 5 to 15 minutes per prompt. Type, tweak, refine, get annoyed when the output lands an inch off the mark, type some more. Multiply that across a week and you’ve quietly burned an afternoon playing part-time prompt engineer.

Here’s the move I keep coming back to. Stop being the prompt engineer. Hand that job to the model.

ChatGPT (and Claude, and Gemini) already know more about what makes a strong prompt than most of us do. So instead of guessing at the magic phrasing, I ask the model to write the prompt for me, then I run that. It’s faster, and what comes back is almost always better than what I’d have hammered out by hand.

The meta-prompt framework

This is the request I use to make the model do the heavy lifting. Drop it into ChatGPT 5, Claude, or whatever you’re running this week.

You are a prompt engineering expert. I need you to create a high-performing prompt that will help me [DESCRIBE YOUR GOAL].

The prompt should:

Be specific and actionable

Include a clear role definition

Provide context and constraints

Specify the desired output format

Include examples where they help

My context: [YOUR ROLE / INDUSTRY]

My goal: [WHAT YOU WANT TO ACHIEVE]

My constraints: [TIME / RESOURCES / REQUIREMENTS]

Create the optimal prompt for this task, then explain why each element improves the output.

Two things make this work. You’re not inventing the structure, the model is, and it knows the structure cold. And that closing line (“explain why each element improves the output”) turns it into a teaching moment. Read the explanation a few times and you start internalizing what a real prompt needs, which sharpens your own writing even on the days you skip the meta-prompt.

This isn’t a shortcut to a clever phrase

Quick caution, because it’s easy to take this the wrong way. The point isn’t to generate a slick one-liner you paste in once and toss. The prompt the model hands back will be a paragraph, not a sentence: a role, context, constraints, an output format. That’s the whole idea.

Treat it like a brief for your sharpest new hire. You wouldn’t hand a new marketer a six-word instruction and expect great work. The meta-prompt is just a fast way to write the kind of detailed brief that actually performs. And then you keep it. Run it again next week. Tune the brackets. A prompt you trust and reuse beats a fresh clever one every single time.

That’s exactly why I tell people the answer isn’t a giant prompt pack. A folder of fifty magic phrases you never run is dead weight. A small set of prompts you’ve tuned and trust is a skill. If you want the longer version of that argument, it’s the core of how I think about AI marketing prompts: build a system, not a collection. The meta-prompt is how I build the pieces of that system without grinding out every one by hand.

Try it this week

Here’s the challenge. Pick ONE marketing task you already do on repeat: ad copy, follow-up emails, research summaries, whatever eats your time.

Run the meta-prompt framework above to generate a better prompt for that task.
Save the result somewhere you’ll find it again.
Test the model-built prompt against your usual approach. Compare the output side by side.

Then keep the winner and run it until it’s boring. That’s the part most people skip, and it’s the part that actually compounds. Routing the right job to the right tool and reusing what works is the whole game, which is what I’m always poking at over in AI for marketing.

Got a meta-prompt you love? Send it over. I’m always testing new ones.

The prompts that survived the test, in your inbox

I’d rather you skip the trial-and-error I do every week. New model ships, I run my prompts through it, and I report back on which ones held up, which quietly broke, and which earned a permanent spot. The newsletter is the shortlist: the prompts that already passed, so you run the winners instead of building the pack. It’s free.

Subscribe Free →