A FREE PROMPT
How to tell when your AI is bluffing.
One paste-in instruction set, built from months of graded AI answers, that makes any assistant show its working before you trust it.
Your email also gets Confidently Wrong: every fortnight, one real case where trusting AI cost someone something, checked against the record and re-run on today’s assistants.
What’s inside
Use. It checks the answer against what you’re about to do with it, not only against the facts.
Scope. It names where the answer is good as of, and stops before it guesses past that.
Filter. Every claim labelled: known, inferred, or a guess. Nothing stated as fact unlabelled.
Risk. It names the one thing that would prove it wrong.
Verdict. A stated confidence level and a usable next step.
I’m Ben. I put the same real questions to six AI assistants and publish every graded answer. The filter is what I run before trusting any of them.