Good Prompting Is Messy
A friend recently told me he was intimidated by the polished prompt examples I share in my posts. He felt he couldn't get good results from an LLM because he didn't know how to write with that clarity and structure.
Truly funny — because my actual prompts are the opposite of that.
I asked ChatGPT to describe my prompt-writing style, and here's what it said:
"Your prompts are usually quick, unpolished brain dumps. You type as you think—often with run-on ideas, minimal punctuation, and occasional spelling errors. They're not carefully crafted instructions but more like raw notes to yourself, where you toss in fragments of context, scattered priorities, and 'I don't know, something like this?' comments. You don't worry about formatting or clarity before hitting enter; you let me figure out what you're trying to accomplish. My first step is often to summarize or restate your request in a structured way before we move forward, so we can align on your goals."
Yup, that's spot-on.
Here's the thing about "good" prompt writing: it's not about writing well. It's about knowing what information to include. Context matters the most. I'll say that again for emphasis. Context matters the most. A messy prompt with the correct details will outperform a beautifully written but vague one every time.
My approach? I dump everything potentially relevant into the prompt — background context, constraints, desired outcomes, and examples. I've already "trained" my chat on tone preferences and format requirements through past writing it has helped me with. I include my uncertainties and my half-formed ideas. Then I let the AI sort through it all and ask for clarification when needed.
The magic isn't in perfect prose — it's in comprehensive information sharing, with you and the AI working through the problem together.
And those clean, structured prompts in my LinkedIn posts aren't my original messy iterations. They're the AI's summary and restatement of what I asked it to do, often after several rounds of back-and-forth clarification.
If anything, this has only become more true. Models keep getting better at inferring what you actually need from messy input. The polished-prompt intimidation is even less warranted than it was.
