Decision support: a second opinion on demand.

Documents are ChatGPT's day job. Decisions are its underrated one. Used with structure, it's a tireless thinking partner that surfaces what you're not seeing — used lazily, it's a yes-man with good grammar. Here's the structure.

01 The trap to avoid first

Ask "should I do X?" and ChatGPT will usually build a supportive case for X — you framed the question, it filled the frame. The fix is never to ask for the verdict first. You want the machine for analysis; the verdict stays yours. Every technique below is a way of getting analysis without contaminating it with your own hopes.

02 The option grid

Start wide: "I need to decide [decision]. Generate the full option space — including the do-nothing option and at least one option I probably haven't considered. For each: strongest case for, strongest case against, what it costs to reverse." That last column is the underrated one — reversible decisions deserve speed, irreversible ones deserve the full treatment below. Use your best model with thinking for this (see the model picker lesson): this is exactly the 'relied on' category.

03 The pre-mortem

The pre-mortem converts vague unease into named risks with tripwires. Half the time you'll proceed anyway — but now you know what you're watching for, and you've pre-committed to noticing it.

04 Assumption hunting and the devil's advocate

05 What stays human

ChatGPT can structure a decision; it cannot want anything for you. Values, appetite for risk, what you'd regret — those inputs are yours, and any answer that arrives without them is generic by definition. The professional pattern: machine builds the grid, machine attacks the plan, you sit with the residue and choose. Then write two sentences about why — future-you, reading that note, is how you actually get better at deciding.