Most of us had to learn how to Google well. It wasn’t just about typing in what you wanted. It was about knowing how to ask the question. Adding quotes, removing keywords, filtering for Stack Overflow. Over time, that skill became second nature. Prompting AI is the same game, just with new rules.
In this first post of the Productive AI series, I want to break down what makes a good prompt, not just for getting something to work, but for working smarter and faster.
Good Prompts Start with Good Structure
When I write a prompt, I don’t just wing it. I usually go in with a clear structure. I think about:
- The role I want the model to take
- The context it needs to understand
- The subject I’m asking about, whether it’s a question, a task, or something else
- The output format I expect
- And what I want it to do if it’s unsure
That last one matters a lot. AI models will sound confident. Always. That’s by design. So if you want to reduce the risk of it confidently hallucinating nonsense, tell it to ask questions before making assumptions. A line like “If anything is unclear, ask before continuing” can go a long way.
Set the Role, Set the Stage
Telling the model who it is helps shape how it thinks. I’ve had great results with prompts like:
“You are in the top 1% of engineering managers. You write clearly, lead with empathy, and prioritize impact.”
That framing alone shifts the response dramatically. I’ve used similar role setups for writing copy, solving home repair problems, navigating legal docs. It works across the board.
Give It What It Needs
The more useful your prompt is upfront, the fewer cycles you’ll need to fix it. If you’re asking for:
- Research or opinions: a well-written question with some context usually works
- Repetitive or structured output: you’ll need to include examples or even a schema (like a sample JSON object)
- Lists or bullet points: tell it not to explain everything unless you want that
Before you hit send, ask:
What can I put in the prompt that will help give me a good answer?
It’s a simple question that forces you to think critically and it’ll save you time.
Format Matters
Want a JSON object? Give an example. Want a plain list? Say “just the list, no extra explanation.” Don’t assume the model knows exactly what you want. Tell it. The more precise you are, the closer the output gets to plug-and-play.
Google Still Has Its Place
Sometimes, the model gets it wrong. Confidently. If I get a weird gut feeling about an answer, I’ll open Google and sanity check it. Especially for niche problems, technical quirks, or anything that smells like a hallucination. Forums and community posts still live in Google. Don’t sleep on that.
Prompting Is a Skill
Just like we had to learn how to search well, we have to learn how to prompt well. It’s a skill. And like any skill, it gets better with practice. The goal is the same: how do I quickly and efficiently get what I need to move forward?
This series is all about putting AI to work in practical ways, not just because it’s shiny, but because it helps us move faster, think better, and get more done.