Using AI for better decisions: frameworks, pros/cons, and devil's advocate
AI is an unusually good sparring partner for decisions — if you stop using it as an answer machine. A workflow for using AI to make better choices, with the prompts that force it to disagree, not agree.
The instinct, when you have a hard decision in front of you, is to ask the model "what should I do?" This is the wrong question. The model will either give you a polite, hedged "it depends" or — worse — a confident answer based on whatever framing you gave it. Neither is decision support; both are confirmation bias dressed up.
There is a better way. AI is an unusually good sparring partner for decisions if you use it that way — to surface arguments you missed, to challenge framings you have settled on, to play the role of the skeptic you wish your team had. This article is the workflow for doing that.
The reframe
The shift in mindset is from "tell me the answer" to "help me think." Specifically:
- Instead of "what should I do?" → "help me understand my own situation."
- Instead of "what are the pros and cons?" → "what are the strongest pros and cons, and which of my assumptions are weakest?"
- Instead of "what would you recommend?" → "what is the most credible argument against what I'm leaning toward?"
The same model produces dramatically different output depending on which question you ask. The reframe is the whole game.
The four-step decision workflow
A repeatable process for any meaningful decision:
- Frame — get the situation onto the table cleanly.
- Generate — produce arguments and options, more than you would alone.
- Stress-test — make the model attack your leaning.
- Decide — pick, calibrated to your real confidence.
We will walk through each.
Step 1: Frame
The single biggest mistake in decision prompts is starting too quickly. You have a feeling about the situation; you describe it briefly; the model latches onto your framing. The output is a polished version of your own initial bias.
Force the model to interview you first.
I am about to make a decision. Before saying anything, ask me 5-7 questions you would need answered to give me useful thinking partnership. Cover: what the actual options are, what my real constraints are, who else is affected, what success looks like, what I'd regret most. Wait for my answers.
Notice the "cover" line. Without it, the model asks generic questions. With it, you get questions that touch each dimension of the decision.
Answer the questions honestly. Often the act of answering reveals something — that you have not actually defined the options, that you have not articulated your real constraint, that you do not know what success looks like. Those revelations are 30% of the value of the whole exercise.
After you answer, the model has the situation. Now it can actually help.
Step 2: Generate
Now ask for arguments and options. The structured-template pattern works well:
Now give me:
>
The three strongest arguments for each option. Quote my answers where they support a point. Mark anything you are unsure about.
>
Two options I might be missing. Sometimes the best choice is one I haven't considered yet. If you see any, name them.
>
The single most important factor I should weigh. What dimension genuinely matters most in this decision?
>
My weakest assumption. Which of the things I told you is most likely to be wrong, or most likely to change?
You will get a list that is broader than what you would have generated alone. The "options I might be missing" line is particularly useful. People often pose decisions as binary ("take the job or not") when there are actually four or five options ("take it, decline, negotiate, take a third option, delay the decision").
A reliable follow-up: "For my weakest assumption — if it turned out to be wrong, how would the decision change?" This is the kind of sensitivity question that humans rarely ask themselves but is exactly what good decision-makers do.
Step 3: Stress-test
Now the most underused move in AI-assisted decisions. Make the model push back.
Now play devil's advocate on my leaning. Assume I am about to choose [option you're leaning toward]. Build the strongest credible argument against it. Be specific — not generic concerns, but the actual scenarios that would make this the wrong choice.
The output is often eye-opening. Models are good at arguing both sides; they just default to whichever side they think you want. Explicitly asking for the counter unlocks the second side.
A more aggressive version:
Imagine it is two years from now and this decision turned out to be a disaster. What happened? Walk me through the most plausible failure scenario, step by step.
This pre-mortem framing is a classic decision-science technique, and the model executes it well. You will often spot risks you missed and find that one of them is more concerning than you assumed.
For decisions where you have a strong intuition, also try:
List three reasons I might be choosing this for the wrong reasons — biases, comfort, sunk-cost, what I want to be true rather than what is true.
Honest self-examination is hard to do alone. The model is happy to do it for you.
Step 4: Decide
By now you have a much richer view of the decision. The final step is to commit — and to commit with calibrated confidence.
Given everything we've discussed, what is your recommendation? Include:
>
1. The choice you would make, in one sentence. 2. Your confidence level: low (40-60%), medium (60-75%), or high (75%+). 3. The single piece of new information that would change your answer.
>
Be direct. Skip the polite "it depends" framing — I want a position.
The confidence level matters. A "high confidence" recommendation is genuinely different from a "this is a close call, slight lean toward X" recommendation, and acting on them should be different. Without an explicit confidence ask, the model will deliver them in the same voice.
Read the recommendation. Then make the decision. The recommendation is input, not output — you are still the decision-maker. But you are now a better-informed one.
A few useful patterns by decision type
Job / career choices. The interview step is critical. Most people frame these as "should I take this job" when the real question is "what do I want from work in the next two years and does this serve that." Force the model to ask about your two-year horizon, your financial constraints, your family situation, and what you would regret most. The decision often becomes clearer once those are on the page.
Hiring decisions. Use the devil's advocate pattern aggressively. Most hiring decisions go wrong because the hiring manager fell in love with one candidate and stopped scrutinising. "Build the case for not hiring [name]" produces useful information.
Strategic / business decisions. Use the pre-mortem ("imagine this failed in two years"). Most strategy decisions are made in optimistic mode; pre-mortem reveals the dependencies and risks that optimism is glossing.
Big purchases. Use the "weakest assumption" framing. Most large purchases assume something about future use or future income that may not hold. Probing the assumption pre-purchase saves real money.
Relationship and personal decisions. Use the "what would I regret most" framing. Personal decisions are dominated by regret in retrospect; surfacing what you would actually regret helps separate it from temporary discomfort.
A common trap to avoid
The trap: using AI as a confirmation machine. You go in with a clear lean, write a prompt that subtly biases toward your lean, and the model dutifully produces support. You feel validated; you decide; nothing has been thought through.
The protection is structural: always run the devil's advocate step, regardless of how clear your lean feels. If after building the strongest counter-argument you still feel confident, you have done real thinking. If the counter-argument unsettles you, that is the model doing exactly what a good adviser would do — pushing back so you do not act on a half-thought.
A useful self-check: at the end of every decision conversation, ask yourself "what did the model say that I did not already think?" If the answer is "nothing," you used the model badly. Restart with a more honest framing.
A Custom GPT to make this reflexive
If you do this often, build a "Decision Sparring Partner" Custom GPT (or Claude Project) with the workflow built in. Instructions like:
When asked about any decision, follow this strict sequence:
>
1. Ask 5-7 questions covering options, constraints, who is affected, success criteria, and what the person would regret most. Wait for answers. 2. List the strongest arguments for each option, options the person may be missing, the most important dimension, and the weakest assumption. 3. Play devil's advocate on the leaning option. Build the strongest credible counter-argument. 4. Give a direct recommendation with a confidence level and the piece of information that would change the answer.
>
Skip steps only if the user explicitly asks to. Push back on rationalisations and biases. Do not give polite both-sides answers — produce positions.
Use this for any meaningful decision. The discipline of running the full workflow on every important choice will compound over months. You will catch a small number of mistakes per year that you would have made without it, and a few of those mistakes will be expensive.
The takeaway
AI is a great decision tool when you stop treating it as an oracle. The hard part of any decision is the thinking — articulating the situation, weighing arguments, finding what you missed, stress-testing your lean. The model accelerates each of those steps. The final choice stays yours.
Frame, generate, stress-test, decide. Four steps, about thirty minutes for a meaningful decision. The decisions get better and the regret rate goes down.