Reading Time: 5 minutes

Hello there, Prompt Lover!

Three weeks ago I did something I'm not sure I'd recommend to anyone.

I sat down with a 200-page academic research paper, a coffee that went cold twice, and the kind of stubborn determination that usually leads to bad decisions.

The paper was called The Prompt Report. Published February 2025. Written by researchers from the University of Maryland, Stanford, Microsoft, Princeton, and OpenAI.

It reviewed 1,565 published papers on prompting and prompt engineering. Identified 58 different prompting techniques. Catalogued 33 vocabulary terms. Ran real benchmarks. Documented a 47-step case study where one researcher spent 20 hours building a single prompt.

I read the whole thing.

Not because I enjoy academic papers. I don't. The citations alone run 40 pages. There's a section with formal mathematical definitions of prompts that made me feel things.

I read it because I had a theory. My theory was this: somewhere in 1,565 papers worth of research, there were prompting techniques that actually work — not "I read a LinkedIn post about this" work, but tested, documented, peer-reviewed work. And I wanted to pull them out, translate them into plain language, and give them to you with a prompt you can copy and use today.

I was right. There were.

What I found changed how I write every prompt. And over the next 31 newsletters, I'm going to give you all of it.

Thirty-one newsletters. One technique per issue. Every prompt tested before it lands in your inbox.

Today we start at the beginning.

Because the most important thing I found in 200 pages of research was also the most basic.

And almost nobody gets it right.

You don’t need to be a coder to make AI work for you. Subscribe to Mindstream and get 200+ proven ideas showing how real people are using ChatGPT, Midjourney, and other tools to earn on the side.

From small wins to full-on ventures, this guide helps you turn AI skills into real results, without the overwhelm.

Get Your Free Guide

When most people write a prompt, they write one thing.

"Write me a marketing email."

"Summarize this document."

"Help me with my bio."

One sentence. One instruction. Then they wait and see what comes back. When it's wrong — and it usually is — they either accept the bad output or they try again with slightly different words.

This is not a prompting problem. It's a structure problem.

The research is clear: a prompt is not a sentence. It's a system. It has components. When components are missing, the AI fills them in itself. And when the AI fills in your missing components, it guesses. Every. Single. Time.

The quality of your output is a direct reflection of how many components you left for AI to guess.

By the end of this issue, you'll have:

• The five components that every working prompt contains

• A template you can use for any task starting today

• A clear explanation of why missing components produce generic output

• The single most important thing the research says about how prompts fail

Let's get started.

Here is a prompt I see constantly:

"Write me a good blog post about AI tools."

It seems fine. It has a topic. It has a format. It says "good," which feels like direction.

It isn't direction. "Good" means nothing to an AI. The AI doesn't know what good looks like to you, your audience, your industry, or your specific situation. It doesn't know if you want 500 words or 2,000.

It doesn't know if you want it conversational or technical, funny or serious, for beginners or professionals. It doesn't know if you want headers or no headers, opinions or just facts, a call to action at the end or not.

So it guesses all of it. And you get the same blog post everyone else gets. Generic. Forgettable. Sounds like a robot wrote it because, structurally, you asked a robot to make every decision.

The gap between what you meant and what the AI understood? That gap is where your results fall apart.

Quick Reality Check

I once counted the missing components in a client's prompt. She had written 47 words. The AI had to make 11 assumptions to produce the output. The output was wrong in 9 of them. She thought the AI was broken. The prompt was.

The research identifies five core components that every effective prompt contains. Not every prompt needs all five at maximum length — but every prompt should account for all five, even if some are brief.

▼ COPY THIS PROMPT template:

Step 1: Copy the template exactly as written above.

Step 2: Fill in the ROLE section first. Be specific. "Expert writer" is weak. "Direct-response copywriter who writes for skeptical B2B founders" is strong.

Step 3: Write your DIRECTIVE as one clear sentence. If you need two sentences, your task might need to be broken into two prompts.

Step 4: Paste everything relevant into CONTEXT. More is better here. Background, constraints, audience details, what you've already tried.

Step 5: Fill in FORMAT before you run the prompt. Decide on length, tone, and structure before the AI does it for you.

Step 6: Add one example in the EXAMPLES section. Even a rough description of the output you want is better than nothing.

Run it. Compare to what you normally get. The difference is the missing components.

The research identifies these five components across thousands of studied prompts. They work because each one removes a decision the AI would otherwise make for itself.

Role removes the identity decision. Without it, the AI picks its own perspective. Usually corporate. Usually vague.

Directive removes the task ambiguity. Without it, the AI interprets what you want. Often wrong.

Context removes the knowledge gap. Without it, the AI assumes. Assumptions are wrong about half the time.

Format removes the structure decision. Without it, the AI defaults. Default outputs look like every other default output.

Examples remove the quality interpretation. Without it, "good" means whatever the AI thinks good means. Which is usually average.

Five components. Five fewer decisions left to chance.

Quick Reality Check

The research paper formally defines a prompt as "an input to a Generative AI model used to guide its output." One sentence. Simple. Which is ironic considering it took them 200 pages and 1,565 citations to fully explain what that means.

Every component you leave out, the AI fills in.

The AI filling things in is not a feature. It's a gap. Sometimes the gap produces something decent. Usually it produces something generic.

The researchers found this across thousands of prompts and dozens of models. The pattern was consistent. Better-structured prompts produced better outputs. Not because the AI got smarter. Because it had fewer gaps to fill.

This applies to every AI tool you use. ChatGPT, Claude, Gemini, all of them. Same structure. Same principle.

The first time you run a five-component prompt, you'll notice the output is different. More specific. More on-voice. Closer to what you actually wanted.

After a week of using this structure for everything, you'll stop going back and forth with AI trying to fix bad outputs. The fix happens before you run the prompt, not after.

I went from averaging four or five back-and-forth revision rounds to getting usable output on the first or second run. Not because my prompts got longer. Because they got more complete.

Pick one task you need to do this week. Anything. Email, post, summary, analysis, draft.

Write the five-component prompt for it before you open your AI tool. Fill in every section. Even briefly.

Then run it. See what comes back. Compare it to what you would have gotten from your old one-sentence version.

That gap is what this series is about closing.

Over the next newsletters, I'm going through every major finding in The Prompt Report. The techniques that are tested and documented. The ones that actually move results.

Here's what's ahead:

Next issue: Why tiny changes in your prompt — a space, a comma, a capital letter — can swing AI accuracy by 80%. This is real research. It's unsettling. And it changes how you build prompts.

Coming soon in the series: the most cited technique in 1,565 papers (it's not what most people think). How to make AI check its own work before you see it. The security problem that's breaking AI tools people trust.

And the 47-step case study that's the most honest piece of writing I've ever seen in a research paper.

Run the five-component prompt on something real this week and reply with what you got.

Tell me what changed. Tell me what still felt off. Send me the prompt if you want me to look at it.

I read every reply.

— Prompt Guy

P.S. This series is based on a 200-page paper called The Prompt Report. I read it so you don't have to. But if you want to, it's free on arXiv. Search "The Prompt Report: A Systematic Survey of Prompt Engineering Techniques." Just maybe have a coffee ready.