AI Guides › Prompting
By Nigel Guy · 2 min read
The default way to ask for help with a choice is to ask for information about the topic — "tell me about pricing models" — when what you actually need is help choosing between two or three specific options you already have in front of you. Asked as a topic, the answer comes back as a broad survey. It's not wrong, exactly, but it doesn't converge on anything, and you're left doing the actual comparison work yourself anyway.
The rule: name the decision and the real options first, and build the prompt around choosing between them — not around the subject they belong to.
| Element | What to put in it |
|---|---|
| The decision | Stated as a forced choice: "should we do A or B", not "tell me about A and B" |
| The options | The actual ones you're weighing — don't let the model invent additional options unless you've explicitly asked it to expand the list |
| The criteria | What matters for this decision specifically — cost, time, reversibility, whatever is actually driving it, named rather than left implicit |
| What happens after | Who acts on the answer and what they do with it — this shapes how much hedging or certainty is actually useful |
Fill in each row before sending the prompt, in that order. A prompt that only fills the first row still reads as a decision on the surface but behaves like an open topic in practice, because there's nothing to weigh the options against.
Skip this structure when you genuinely want open exploration and haven't yet identified real options — forcing a decision frame too early produces false confidence in a shortlist you haven't actually found the full range for. Skip asking for a recommendation without naming real criteria; without them, the model will supply generic ones of its own, and a recommendation built on invented criteria isn't worth much.