From The No-Hype AI Playbook

Chapter 3 — Garbage in, garbage out: the five-minute fix

The resurrection

Think of the worst answer an AI ever gave you — the one that sent you away convinced. (Never used one? Borrow Frank's: "tell me about retirement planning.") Now rebuild it with the four parts: who you are, what's going on, what you actually want, what shape the answer should take. Type the new version into your tool from chapter 2 and compare what comes back against what you remember.

That distance — beige to bespoke — was never the machine's range. It was yours. It still is, every time you type.

The briefing, three ways

1. The fill-in-the-blanks briefing

Keep this skeleton somewhere handy until it's habit

I'm [who you are, one sentence]. Here's my situation: [what's going on — the over-the-fence version]. What I want is [the actual want, not the topic]. Give it to me as [steps / options with pros and cons / a plain-English explanation], and keep in mind [budget, deadline, bad knee, 'keep it simple'].

2. The interview opener

For when you don't know which details matter

I want help with [the thing], but I'm not sure what information you need from me. Before you answer, ask me your questions one at a time. When you have enough, say so, then give me [the shape you want].

3. The steering wheel

Not a prompt — a phrasebook for talking back. Use freely, no pleasantries required

“Shorter.”
“Plainer — explain it like I've never done this.”
“What's the cheaper version?”
“Give me the same thing as a checklist.”
“You're assuming I have a truck. I don't.”
“Skip the warnings, I've been doing this for forty years.”
“Start over, but this time ask me questions first.”

That last one is the escape hatch for any conversation that's gone sideways. There's always an escape hatch. It's one of the few machines in your life for which that is reliably true.

Works in ChatGPT, Claude, or any of their cousins.