TL;DR
- You don't need to be technical to prompt well. You need to stop typing three words and hitting enter.
- The one habit that fixes almost everything is context: tell the AI who it should be, what you want, and what a good answer looks like.
- Seven simple ways of asking cover most of daily life, from summarizing an email to planning a week of meals.
- The best place to practice is your own life: your messages, your plans, your notes.
This piece is from Code Meet AI, my newsletter where I document how I actually use AI to ship real things, receipts included. New guides land there first.
Prompting is just how you ask an AI for something. The AI reads your words, not your mind, so a lazy question gets a lazy answer. The fix is context: tell it who you want it to be, what you actually need, and what a good result would look like. After that, a handful of simple patterns handle almost anything. No jargon. You get good by practising on the things you already do every day.
The AI reads your words, not your mind
The first time a lot of people try one of these tools, they type something like "make me 1 million," read the bland thing that comes back, and quietly decide it is overrated. I did a version of this for months. I treated it like a smarter Google, typed three words, and was annoyed when i got three-word-quality answers back.
Spoiler alert: the "make me 1 million" prompt doesn't work. Which i think you already know by now.
What changed for me was not a better app. It was an afternoon in a park. Instead of firing one-line questions at it, i opened the AI on my phone and just talked to it for about three hours: what i was working on, what was stuck, what i had been avoiding. By the evening it could help me in a way it never could when i was firing one-liners at it.
That is the whole game. When you leave a question vague, the model fills the empty space with something generic, because generic is the safest guess when it does not know what you actually meant.
What a good prompt actually contains
A good prompt is not longer; it is clearer. Four things usually do it, and you rarely need all four: who you want the AI to be, what you want, the bit of context it cannot see, and what a good answer looks like.
Lazy: "give me dinner ideas." You get a generic list, half of which you would never make.
Clear: "You're a practical home cook. Give me three vegetarian dinners for a tired weeknight. I have pasta, spinach, and eggs, i want each under twenty minutes, and just give me the three with a one-line reason, no essay." Same question, but now it knows the situation, the limits, and the shape you want back.
The same question, minus the guesswork.
The seven ways to prompt
Friendly names first, the technical name in brackets for the curious.
- Just ask it clearly (zero-shot). "Summarize this email in three bullets and tell me whether i need to reply." Your default for anything simple.
- Show it an example (few-shot). "Here's a thank-you note i wrote once that felt right [paste it]. Write one like it for my neighbour." It copies your taste instead of guessing at it.
- Make it think out loud (chain-of-thought). "I'm choosing between two flats. Walk through commute, cost, and daylight step by step, then tell me which one and why." Forcing the thinking first beats a snap answer.
- Break the big thing into steps (decomposition). Planning a five-day trip becomes: rough day-by-day, then one day in detail, then the packing list. Small steps you can check.
- Ask for the plan, then the work (reason then act). "Tell me how you'd structure a monthly budget for an irregular income, then build it as a table i can fill in."
- Have it check its own answer (self-critique). "Reread that message to my landlord and flag anything passive-aggressive, then give me a warmer version." The easiest quality check there is.
- Keep improving, and save what works (iterate). Tell it exactly what was off, teach it your taste, and save the asks that work so you can reuse them.
If you only remember three: ask clearly for the simple stuff, make it think out loud for anything you are deciding, and have it check itself before you send anything that matters.
The seven, one line each. This is the cheat sheet.
Your own notes are the best place to practice
You learn this on your own life, not from a course. I keep what people call a second brain, a plain folder of my own notes that the AI can read, and it is the best practice ground i have.
Try the smallest version. Drop a week of messy notes into one place, then ask: "group these into themes, tell me what i keep circling back to, and what i seem to be avoiding." Giving it your real material instead of a blank question is patterns 1 through 7 at once. If you want the full setup, i wrote up how i built mine with Obsidian.
Ask, read what came back, steer, save what works. Then again.
Where a better question stops helping
Asking well fixes one conversation. It does nothing about the fact that the AI forgets you the moment you close the chat. That is where the work moves from the question to what is around it: giving the AI a memory it can read, which is exactly what a second brain is for. That is the next part of this series.
How to get good: practise on your own life
Try, look at what came back, adjust, go again. When an answer is bad, read it instead of deleting it, because it is telling you which of the four things you left out. And when a way of asking works, hand it back to the AI: "write me a cleaner version of this prompt i can reuse." Using the AI to improve how you talk to the AI is the fastest way i know to get good.
The seven live on a one-page cheat sheet, plain language, ready to copy. Grab it free here.
I'm Malik. I build with AI every day and write about using it well, minus the hype. This is part 01 of Using AI, properly. Next: giving the AI a memory. New parts land first at codemeetai.substack.com.
Top comments (0)