Say What You Mean: How to Write Better AI Prompts
Better prompts come from better writing: say who the output is for and why, give the context the model can't guess, show an example, name the format, say what to avoid, ask for steps on complex tasks, and treat the first answer as a draft — then check any facts before you use them.
"Prompt engineering" sounds like it needs a lab coat. Mostly it needs the skill this whole site is built on: choosing words carefully. A language model can only work with what you give it, and it will fill every gap you leave with a guess.
Here are seven habits that close those gaps — starting with a before-and-after, because an example beats an explanation.
Before and after
Before: "Write some cat puns."
After: "I'm making a birthday card for my sister, who's a vet. Write 10 cat puns she'd enjoy — family-friendly, under 12 words each, ideally with an animal-care angle. Skip 'purr-fect'; she's heard it. Give them as a numbered list."
The second prompt isn't clever. It just answers the questions the first one left open: who it's for, what it's for, how long, what to avoid and what shape the answer should take. Every habit below is a version of that.
1. Say who it's for and what it's for
"Explain RAG" gets a generic answer. "Explain RAG to a marketing team deciding whether to buy an AI support tool" gets a useful one. Audience and purpose change what a good answer is, and the model can't know either unless you say.
2. Give it the context it can't guess
The model knows nothing about your situation, your company or the document open on your other screen. If the answer depends on it, paste it in. Most disappointing answers are really missing-context answers.
3. Show what good looks like
One example of the output you want is worth several adjectives. "Punchy" and "professional" mean different things to different people; a sample paragraph means one thing.
4. Name the format
A list, a table, three options, under 100 words, a subject line and a two-sentence email — say it. Format instructions are the easiest win in prompting, and the most often skipped.
5. Say what to avoid
Constraints are instructions too: no jargon, don't mention pricing, skip the obvious pun. Telling a model what you don't want often improves the result more than adding what you do.
6. For complex tasks, ask for steps
For anything with several moving parts — a plan, a comparison, a calculation — ask the model to outline its approach or work through the steps before giving a final answer. It makes errors easier for you to spot, even when the model does some of that reasoning on its own.
7. Treat the first answer as a draft
The first response is a starting point. Say what to change: shorter, warmer, fewer adjectives, a different second point. You can also ask the model to put questions to you before it starts — it will often surface the context you forgot to give.
And whatever it produces, check the facts. Models can state wrong names, numbers and quotes with total confidence; here's why that happens and one common fix.
Keep reading
Frequently Asked Questions
What is prompt engineering?
It's the practice of wording instructions to an AI model so it produces useful output. In practice it's mostly clear writing: purpose, audience, context, an example, a format and constraints.
Do longer prompts work better?
Not automatically. Relevant detail helps and padding doesn't. A short prompt with the right context will beat a long, vague one.
Why does AI give different answers to the same prompt?
Language models generate text with a degree of randomness, so the same prompt can produce different wording or ideas each time. Settings such as temperature control how much variation you get.