Good answers aren't luck. They're technique.
The same tool gives one person gold and the next person mush. The difference is almost never the model — it's how you ask. Here's how to ask, for everyday work and for writing code.
A good prompt has four parts
Vague in, vague out. Name who it's for, what you want, the situation, and the shape of the answer — and the quality jumps immediately.
“Write something about email marketing” gets you filler. “Write a 5-bullet summary of why small clinics should use email marketing, for a non-technical owner, one sentence each” gets you something you can send. Same model. Thirty extra seconds.
Ready-to-use prompts
Steal these. Tap a category to narrow it down, then copy the one you need. Replace anything in [brackets].
Techniques that actually move the needle
Not tricks. Habits. Each one reduces the ambiguity the model has to guess at — which is the whole game.
Be specific about “done”
Name the task, who the output is for, and what a finished answer looks like. The single biggest upgrade most people can make.
Show one example
Paste a sample of exactly the kind of output you want. One good example teaches the model more than a paragraph of instructions.
Give it a role
“You are a seasoned [X].” This one line sets tone, depth, and vocabulary — one of the most reliable ways to steer the answer.
Tell it to think first
For anything with reasoning, add “think it through step by step before answering.” Working out the steps before committing measurably improves accuracy on hard problems.
Hand it the source
Don’t rely on its memory — paste the document, the data, the email thread. Grounding it in real material is what stops it from inventing things.
Specify the format
Length, structure, bullets vs. prose, a table, a subject line. Each constraint is one less thing you have to fix afterward.
Say what NOT to do
“No jargon. Don’t invent statistics. Under 150 words.” Negative constraints are as powerful as positive ones.
Treat the first answer as a draft
The real skill is the follow-up. “Shorter.” “More concrete.” “Now make the counter-argument.” It’s a conversation, not a vending machine.
Make it interview you
“Ask me what you need to know before you start.” The model often knows what’s missing — let it pull the context out of you.
Separate instructions from content
On long prompts, label the parts — put your instructions in one block and the material in another.
Chain big jobs into steps
Outline, then draft, then tighten — each step in its own prompt. Small focused passes beat one giant request every time.
Trust words. Verify facts. Never paste people.
Lean on it
- Summarize & explainIts home turf
- Rewrite the toneWarmer, clearer, shorter
- Draft & templateFirst drafts, not final ones
- Brainstorm & outlineOptions and angles
Always double-check
- Citations & casesIt invents real-looking ones
- Specific numbersRates, doses, totals
- Anything recentIt may be on old data
- MathIt's a writer, not a calculator
The free versions of ChatGPT and Gemini aren't built to legally hold protected or proprietary information. The fix isn't “don't use AI.” It's using AI that keeps the data inside your building. That's what Piazza builds.