Build a better prompt, copy it, run it anywhere.
Type a question, choose how the AI should handle it, copy the prompt. This builds the prompt; it doesn’t run it.
Use only the information I give you here. Don't fill in gaps from memory; if something you need isn't here, say so. If a question is past what you can confirm, or the figure is the sort that changes, say you're not sure rather than guess.
Tag each claim as you make it: [sourced] with where it came from, [inferred] if you're reading between the lines, or [guessed] if you're filling a gap.
Name the biggest risk if I act on this and you're wrong.
End with one clear answer and a confidence level (low / medium / high).
Should I take ibuprofen with my blood pressure medication?
Then check the answer you get back
The prompt only does half the job. When your AI answers, run these checks on what it says. A good Prompt Stack answer earns a yes on each one.
- Did it stick to what you gave it, or pad the answer from memory?
- Did it label its claims (sourced / inferred / guessed), or blur them into one smooth answer?
- Did it say where it was unsure, or quietly guess and move on?
- Did it name a real risk and the worst case, or just hedge?
- Did it commit to one answer with a confidence level, or sit on the fence?
- Check any figure, date, name or citation against the source yourself.
Want the why behind the four stages?
This builder is the Prompt Stack made into buttons. The full method, with each stage worked through on everyday decisions and on real money, is one page over.
Read the full method →Confidently Wrong, my fortnightly newsletter: I run the big AI assistants through the same checkable questions and send you the receipts when one quietly gets it wrong, plus the fix worth keeping. A quiet sense of humour about the whole business.
One email a fortnight. Leave whenever you like.