How to Write Prompts That Actually Get You What You Want

The three things every good prompt needs

Most vague prompts are missing one or more of these:

• The task — what, specifically, do you want done? "Help with X" is a topic, not a task. "Summarize X in 3 bullet points" is a task.
• The audience — who is this for? A prompt written for a 10-year-old and one written for an executive should produce very different answers, but the AI can't guess which you want unless you say.
• The format — how long, and in what shape? One paragraph vs. a bulleted list vs. a single sentence changes everything about how the answer gets structured.

A before-and-after example

The second version tells the model exactly what to focus on (a plain-language explanation, not the math), how long to make it (one paragraph, not an essay), and who it's for (someone who needs business impact, not algorithm details). Same topic, dramatically more useful answer.

Turning a vague request into a specific one

A reliable technique: take a vague request and ask yourself "what would I tell a new employee if I handed them this task with zero context?" Whatever you'd naturally clarify — the audience, the constraints, the goal — belongs in the prompt.

Why this matters more as tasks get harder

For a one-line factual question, vague prompts mostly work fine. The gap widens fast on anything with judgment involved — writing, planning, analysis — where "good" depends entirely on constraints the model can't read your mind about.