What Google's boffins taught me about writing AI prompts

When I started using AI, I wasted hours trying to get Claude and ChatGPT to generate a copy.

My prompts were all over the place:

“Write me compelling sales copy for a coaching program for content creators.”
“Make my sales copy more persuasive.”
“Improve my copy so it sounds better”

The outputs?

Generic, bland, and useless.

I wanted to set AI aside and write all my content without help, but AI isn’t going anywhere. It is better to learn how to use it or get left behind.

So, one morning, I waded through a massive document about AI from the boffins at Google. I discovered a framework that expert prompt engineers use.

Break your prompts into four essential parts:

Persona: Who do you want the AI to be, e.g., Steve Jobs, James Patterson, Gary Halbert
​Task: What specific action do you want?
​Context: Critical background information
​Format: How you want the output structured (upload training data if you have it)

Here’s what happened when I rewrote my prompt using this framework:

“You’re a direct response copywriter who studied under Gary Halbert. Write a sales email for my $997 coaching program that helps content creators earn their first $5K. Use story-driven copy with a clear call to action. Format it with a hook, story, offer, and P.S.”

The difference? Night and day.

Instead of vague marketing speak, I get a compelling draft for my target audience. I still insert some personal stories and insights that AI can’t generate.

Think of AI as your writing assistant.

Here’s why this simple prompt engineering framework works:

Persona gives the AI a specific role to embody. It’s the difference between asking “anyone” for advice versus asking an expert like copywriter Gary Halbert in my above prompt.

Task clarifies precisely what you want - not just “write copy” but “write a sales email for a specific product so we can do x, y or z.”

Context provides the crucial background that shapes the output. Without it, you’re asking the AI to guess what you want. It’s useful, but it’s not a mind-reader. And without context, it’ll spit out boring content.

Format ensures you get the deliverable in a structure you can use. Building a library of training data is key here.

Think of prompt engineering as like baking:

You need the right chef (persona), clear instructions (task), quality ingredients (context), and proper measurements (format).

Miss any of these, and your cake falls flat just like my early prompts did.

Want to test this framework yourself?

Pick something you’ve been struggling to create. Maybe it’s website copy, email sequences, or your social posts.

Rewrite your prompt using all four parts. Then, compare the outputs.

I bet you’ll see an immediate improvement in quality and usability.

I didn’t stop there, though. I built out the prompt further by iterating it, as you can see below.

I could go on but, if you need help mastering prompt engineering, reply “ENGINEER” to this email. I’m working with 5 content creators who want to master AI for their business.