The core idea

Three questions.
Every time. Forever.

You don't need to learn "prompt engineering." You don't need a course. You need three small questions that, over about a week, become a habit. Once they are, you'll catch 80% of the mistakes AI makes — and get answers worth trusting on the rest.

The Three Questions

Read these once. Come back to this page whenever you need the refresher.

01

What exactly do I want?

Vague in, vague out. Most disappointment with AI is the user's fault, not the AI's. If you ask "tell me about leases" you'll get a Wikipedia article. If you ask "I'm renewing a lease, the landlord added three clauses I don't understand — walk me through what changed and flag anything I should push back on" — now it has a chance. Write to AI like it's a brilliant, slightly distracted intern who has never met you.

02

What would "correct" even look like?

Before you hit send, ask yourself: how would I check this? What would a good answer look like? If the answer is "I'd know it when I see it" — fine, but be on guard. If the answer is "I could look it up, call someone, do the math" — great. Now you have a verification plan. If the answer is "I have no idea what 'right' would look like" — that is not the question you should be asking AI alone. Or at least, not without a human expert as the final check.

03

What can't it know?

Four things AI fundamentally can't know: what happened after its training cutoff (it doesn't know today's news); anything not in its training data (your town's quirks, internal documents, rare topics); anything private about you (your health, your finances, your family, unless you told it in this conversation); and whether it's right (it has no compass for truth — it will be wrong the same way it's right, with full confidence). Before trusting an answer, say out loud: "what doesn't it know here?"

Why three questions and not ten?

Because ten questions is a course. Three is a habit. The goal isn't to make you a prompt engineer. It's to give you something small enough to actually use — every time, forever, with any AI tool that gets invented next year or the year after.

Try them on something real

A task you've been avoiding. A document you don't understand. A decision that's been stuck. Type it in, walk through the three questions, get back a prompt you can paste straight into ChatGPT.

The Prompt Builder

Turn a rough request into a useful prompt

Nothing leaves your browser. No account needed.

0
The task (one sentence — rough is fine)
"Understand this bill." "Write this email." "Explain this news article to me."
1
What do you actually want?
The who, the what, the why. Like you're briefing an intern.
2
What would a good answer look like?
How will you check it? What would "correct" sound like?
3
What can't it know?
Your specific situation, your insurance, your state, your provider — whatever isn't in the conversation yet.
Your prompt
Before you send: never paste anything private you wouldn't email a stranger — social security numbers, account numbers, your kids' full names + school, full login info. AI works just fine with "[NAME]" and "[ADDRESS]".

Two bonus habits

After a week with the three questions, these take you the rest of the way.

Next: try it tonight

Pick one real thing before bed. A bill, an email, a document. Use the three questions. Verify one thing. That's the whole assignment.

See real examples FAQ & reframes