How to Write Better AI Prompts (With a Free Checklist)

Ask ChatGPT to "write something detailed about our product launch" and you will get exactly what you asked for: something. Detailed, sure, but detailed about what, in what tone, for whom, how long? The model isn't being lazy — it's filling in every gap you left with its own best guess, and its best guess is rarely the one you had in mind.

The fix isn't a magic phrase or a prompt-engineering course. It's five specific pieces of information that, when present, consistently produce sharper replies — and when missing, consistently produce the kind of generic mush that makes people say AI "isn't that smart." It's smart enough. It just answered the question you actually asked.

The five things every strong prompt has

• A role — who should the model act as? "You are an expert copy editor" changes vocabulary and judgment, not just tone.
• A task — a plain instruction verb. Write, summarize, compare, extract. Not "help me with," which could mean anything.
• An output format — bullet points, a table, JSON, a short paragraph. Say the shape you want, not just the content.
• A length — "under 150 words" or "three sentences." Without it, you get whatever length felt average in the training data.
• A tone or audience — "for a beginner," "formal tone." The model can write for an expert or a ten-year-old equally well; it just needs to know which.

Photo by Kaitlyn Baker on Unsplash

Notice none of these require you to know any AI jargon. You already do this instinctively when briefing a person — you don't hand a junior designer "make something detailed" and walk away. You tell them who it's for, how long, what format the client expects. Prompts are briefs. Treat them like one.

The mistakes that aren't about missing information

Beyond the five gaps, a few habits actively work against you even in a prompt that looks complete on the surface.

• Contradictions. "Give me a brief but extremely detailed report" asks the model to guess which instruction wins. It will pick one — probably not the one you meant.
• Vague intensifiers with no number behind them. "Comprehensive," "thorough," and "detailed" all mean nothing on their own. Comprehensive compared to what — a tweet, or a thesis?
• Pronoun soup. "Take this and fix it, then check it again and rewrite it so it reads better than it did." By the third "it," even a human reader has to stop and figure out what you mean.
• Politeness padding. A single "please" costs nothing. Three "if possible, kindly, please" phrases in one prompt cost real tokens and change nothing about the answer.

A before-and-after

“Before: "Write something detailed about our new product launch for social media."”

This prompt has a task (write) but nothing else. No role, no format, no length, no audience. The model has to invent a platform, a tone, and a length from nothing — and it will, confidently, in whatever direction its training data leans.

“After: "You are a social media copywriter for a B2B SaaS company. Write an Instagram caption announcing our new product launch, under 40 words, in an upbeat but professional tone, ending with a call to action."”

Same task, same product, same model — but every gap is now filled with a decision you made on purpose instead of one the model made for you. That is the entire difference between a first draft you throw away and one you can actually use.

Photo by Trey Gibson on Unsplash

Turn this into a habit, not a one-off

The honest way to build this habit is to run your prompt through the same five-point check every time until it's automatic — the same way a good headline gets checked against a formula before it ships. You don't need to memorize the checklist forever. You need to run it often enough that you start writing prompts that already pass it.

Paste a prompt you're about to send, see the clarity score climb as you add the missing pieces, and copy the restructured version once every section is filled in. It takes less time than reading this article did.