The Decision Brief Prompt: Turn Any Messy Call Into a Scored Recommendation

What you'll learn in this article:

• Why most decision prompts return confident-sounding mush
• The exact Decision Brief prompt, ready to paste
• How the confidence-level and missing-information lines change the output
• Three high-stakes decisions where this earns its keep
• How to push back when the recommendation feels wrong

Time to value: 2 minutes.

Hard decisions stall in the same place every time. The options are clear enough, but they live in your head as a fuzzy list, and turning them into something you can defend means building a comparison doc you do not have time for. So the decision drifts, or you make it on gut and hope.

Asking ChatGPT for help usually makes it worse, not better. A vague prompt like "should I build or buy this" returns a balanced-sounding paragraph that commits to nothing and teaches you nothing. The model sounds confident because it always sounds confident, which is exactly the trap.

The Decision Brief prompt fixes this by forcing structure: scored options, one recommendation, a stated confidence level, and the single piece of missing information that would most change the answer. That last element is the whole trick, and it is what turns a generic answer into a genuinely useful brief.

The prompt

Paste this into ChatGPT (Plus or Go, running GPT-5.2). Fill the two bracketed fields with your actual decision and context. Everything else stays as written.

DECISION BRIEF PROMPT

Act as a sharp operating advisor. I have to decide: [describe the decision in 2-3 sentences]. Here is the context: [paste the facts, constraints, and budget]. Produce: (1) the 3 strongest options, (2) a tradeoff table scoring each on cost, speed, and risk from 1 to 5, (3) your single recommendation, (4) your confidence level as a percentage, and (5) the one piece of missing information that would most change your answer. Keep it under 300 words.

Why each part earns its place

The structure is not decoration. Each numbered element forces the model out of a comfortable failure mode.

Asking for three options stops it from rubber-stamping the one you already named. Requiring a scored tradeoff table on cost, speed, and risk forces a real comparison instead of prose that gestures at tradeoffs. Demanding a single recommendation removes the "it depends" escape hatch. The confidence percentage is the honesty check: a recommendation at 55 percent confidence is a very different signal than one at 90 percent, and the model will only tell you the difference if you ask.

The fifth element, the one piece of missing information, is the most valuable. It surfaces the real risk in the decision, the thing you have been avoiding looking at. Often it names the exact data point that, once you go find it, makes the decision obvious. That single line is worth more than the recommendation itself.

A worked example

Say the decision is whether to hire a second salesperson now or wait two quarters. You paste your pipeline numbers, current rep's capacity, and cash position. The brief comes back with three options (hire now, hire in Q2, hire a contractor), a scored table, a recommendation to wait one quarter at 65 percent confidence, and a missing-information line: "Your current rep's close rate on inbound versus outbound, which determines whether a new hire would have enough qualified pipeline to ramp." That line tells you exactly what to check before committing to a salary.

How to push back

The brief is a starting point, not an oracle. When the recommendation feels wrong, do not argue in vague terms. Feed it the reason:

CHALLENGE FOLLOW-UP

Your recommendation assumes [the thing you disagree with]. That assumption is wrong because [your reason]. Redo the brief with the corrected assumption and tell me if your recommendation changes and by how much your confidence shifts.

This keeps the analysis rigorous instead of letting it collapse into agreement. A good brief should hold its position when your pushback is weak and change cleanly when your pushback is strong.

Where it earns its keep

Use this for decisions where a wrong call is expensive and the data is incomplete: build versus buy, vendor selection, and hiring tradeoffs. These are the calls that genuinely benefit from forced structure, because the cost of drifting or guessing is high. For low-stakes, reversible choices, a quick gut call is fine, do not over-engineer a lunch order.

Action Steps Summary

• Paste the prompt: Drop the Decision Brief prompt into ChatGPT and fill the decision and context fields.
• Read the confidence number first: A 55 percent recommendation is a flag to gather more data; a 90 percent one is a near-clear call.
• Act on the missing-information line: Go find the one data point it names before you commit budget.
• Push back with reasons: Use the challenge follow-up to test whether the recommendation holds.
• Reserve it for high-stakes calls: Use it on build-vs-buy, vendor, and hiring decisions, not reversible small ones.