← back
#ai #productivity

How to Write Prompts That Actually Get You What You Want

10 min read · devnotes

Same AI, wildly different results — the only variable is how you asked. Most people get mediocre AI output not because the model is weak, but because the request was vague. Here's how to be specific and assertive about what you actually want, so you stop re-rolling the same question five times.

Vague prompts get vague answers

"Write a blog post about productivity" and "write a practical blog post for overwhelmed freelancers about cutting task-switching, using two simple systems they can start today" ask for the same general thing — but only one of them tells the AI enough to actually be useful. The AI isn't guessing what you meant; it's filling gaps you left open, usually with the most generic, average version of an answer.

The four things worth stating explicitly

ElementWhat it answers
RoleWho should the AI act as? ("You're a senior copywriter," "act as a strict code reviewer")
ContextWhat's the situation? Who's the audience, what's already been tried, what constraints exist?
TaskThe specific action, stated as a clear instruction, not a topic
FormatHow should the output look? Length, structure, tone, bullet points vs. prose

You don't need all four in every single prompt — but for anything that matters, filling in at least role, context, and format turns a generic answer into something close to usable on the first try.

See the difference in practice

✕ weak prompt: "Give me feedback on this email" ✓ prompt with role, context, format: "You're a direct, experienced manager. I'm about to send this email to push back on a deadline my team can't hit. Give me blunt feedback on tone and clarity, then rewrite it in under 100 words, firm but professional."

The second version doesn't just ask for "feedback" — it defines the lens to give feedback through, the stakes, and exactly what a good output looks like.

Give it examples when the format matters

If you have a specific style or structure in mind, showing one or two examples ("few-shot prompting") almost always beats describing it in words. AI is far better at matching a pattern you show it than following an abstract description of that pattern.

"Write 3 more product taglines in this exact style: - 'Coffee that respects your morning.' - 'Sleep, but make it science.' Same tone: short, a little clever, no exclamation points."
AD SLOT — 728×90 or responsive (mid-article)

Ask it to think in steps for anything complex

For reasoning-heavy tasks — debugging, math, multi-part decisions — explicitly asking the AI to work through the problem step by step before giving a final answer noticeably improves accuracy. It's the difference between asking for a conclusion and asking for the reasoning that leads there.

"Walk through this step by step before giving your final answer: should I refactor this function now, or ship it as-is and fix it next sprint? Consider time cost, risk, and what breaks if I wait."

Treat the first response as a draft, not the answer

The people who get consistently good results rarely accept the first output. They read it, then respond with something like "make this more concise," "the tone is too formal, loosen it up," or "this misses X — redo it accounting for that." Iterating two or three times, in the same conversation, almost always beats trying to write one perfect prompt from scratch.

quick tip
Being assertive doesn't mean being rude — it means not leaving important decisions up to the AI's guesswork. "Make it good" leaves everything open. "Keep it under 150 words, cut the jargon, and end with a clear next step" leaves nothing open.

Build a small library of prompts that already work

Once you find a phrasing that reliably gets a good result for something you do often — summarizing meeting notes, drafting outreach messages, reviewing code — save it. Reuse and tweak it instead of starting from scratch every time. This is exactly how the lessons and prompt examples throughout our other AI articles were built: not one perfect prompt, but a small set of reliable ones, reused deliberately.

bottom line
State the role, the context, and the format you want, show an example if the style matters, ask for step-by-step reasoning on anything complex, and treat the first answer as round one, not the final draft. That combination gets you dramatically better results than the same request typed vaguely.
AD SLOT — 300×250 or responsive (in-article)