Here's a pattern worth noticing: two people use the same AI tool, ask about the same thing, and get wildly different results. One gets a vague, generic paragraph. The other gets exactly what they needed. The model didn't change between them — the prompt did.
A prompt is just the instruction you give an AI. Think of it like a brief you'd hand a sharp new assistant: the more clearly you say what you want, who it's for, and what "good" looks like, the better the work comes back. None of these habits require technical skill. They're small changes to how you ask — and they work on any AI assistant. Front-loaded strongest first.
1. Give the AI a role before you give it a task
Start by telling the model who it should be. "You are a patient math tutor explaining to a 12-year-old" produces a different answer than "You are a research mathematician." This is called role prompting, and it's the highest-leverage habit on this list because it sets the vocabulary, depth, and tone for everything that follows.
Compare "Explain compound interest" with "You're a financial educator. Explain compound interest to someone who's never invested, using one everyday example." The second gets you a usable explanation; the first gets you a textbook definition. One sentence of role does most of the work.
2. Show the format you want — don't just describe it
If you want a table, paste a tiny example table. If you want three bullet points, say "exactly three bullets." Models are extremely good at pattern-matching, which means copying the shape of an example you give them. Describing a format in words is fine; showing it is far more reliable.
Try ending a prompt with: "Format your answer as: Claim — one-sentence reason." You'll get clean, consistent output instead of a wall of prose you have to reorganize yourself.
3. Give it the context it can't guess
The AI doesn't know your situation unless you tell it. "Write a follow-up email" is a guess; "Write a short follow-up email to a client who hasn't replied in a week — friendly, not pushy, and remind them the proposal expires Friday" is a brief. The relevant facts — audience, constraints, deadline, what's already been tried — are exactly what turns a generic answer into yours.
A simple test: before you hit send, ask yourself, "Could a stranger write a good answer from this prompt alone?" If not, the missing piece is context.
4. Ask for the thinking, not just the conclusion
When a question has any reasoning to it, add "explain your reasoning step by step" or "show your work before giving the answer." Walking through the steps tends to make the final answer more accurate — and just as important, it lets you spot where the logic went wrong instead of trusting a conclusion you can't check.
This is the difference between "Is this contract clause risky?" and "Read this clause, list what each part means, then tell me where the risk is." The second is auditable. You can see the chain, not just the verdict.
5. Constrain the length and the audience
"Explain quantum computing" can mean a tweet or a textbook — so the model picks for you, usually badly. Pin it down: "in two sentences," "in under 100 words," "so a high-schooler gets it," "for an expert who already knows the basics." Length and audience together control how much the model assumes you already know.
A handy phrase: "Keep it to a paragraph a busy person would actually read." Vague to a human, but it reliably nudges the model toward tight, jargon-light writing.
6. Iterate — treat the first answer as a draft
The single biggest mistake is treating the first response as final. It's a draft. Reply with "make it shorter," "you missed the cost angle," "rewrite #2 to be more skeptical," or "good — now give me the opposite view." The AI keeps the conversation context (everything said so far in the chat), so each follow-up builds on the last instead of starting over.
Strong users rarely get there in one prompt. They get there in three. The skill isn't writing the perfect first question — it's steering.
7. Ask it to ask you
When you're not sure how to frame a request, hand the work back: "Before you answer, ask me up to three questions that would help you give a better response." This flips the model into surfacing the missing context you didn't think to provide — the deadline you forgot to mention, the audience you assumed it knew. It's the fastest way to fix a prompt you couldn't have fixed yourself.
Where to start today
Pick habit #1 — give the AI a role. It takes one extra sentence and changes the quality of every answer after it. Add the others as they become second nature.
One more thing: the best prompts are rarely written from scratch. The ones that work, you'll want to reuse. Keep a running note of your sharpest prompts — the role lines, the format templates, the follow-ups that always land — so you're refining a personal library instead of reinventing the question every time. In JustJot.ai, that's a single note you can recall in seconds with semantic search: type what you're trying to do, and the prompt that worked last time comes back to you.
A good prompt is just a clear question. These habits are how you learn to ask one.