Using AI for Decision Making — Without Getting Yes-Answers

Published March 7, 2026 · 11 min read

You're facing a big decision. You open your AI chatbot. You describe the situation. You already know which way you're leaning. The AI confirms your instinct. You feel confident. You proceed.

Congratulations — you've just used the most powerful analytical tool in history as a confirmation machine.

This is how most people use AI for decisions. And it's worse than not using AI at all, because it gives you false confidence. At least without AI, you'd know your decision was based on gut instinct alone. With AI confirmation, you think it's been analytically validated. It hasn't. It's been analytically decorated.

How Most People Use AI for Decisions (Wrong)

The typical pattern looks like this:

1. You have a decision to make
2. You already have a preference (consciously or not)
3. You describe the situation to AI, framing it through your existing perspective
4. The AI picks up on your framing and provides analysis that supports it
5. You feel validated and proceed with increased confidence

The problem isn't step 1, 2, or 5. It's steps 3 and 4. Your framing biases the AI's analysis, and the AI's training bias toward agreement amplifies your framing. The result is a confirmation feedback loop that feels like rigorous analysis but is actually just your own bias reflected back at you with better vocabulary.

The Confirmation Bias Trap

Confirmation bias is the tendency to seek, interpret, and remember information that confirms what you already believe. It's the most well-documented cognitive bias in psychology, and it's devastating in decision-making.

AI makes confirmation bias worse, not better, for three reasons:

• AI is articulate: It can construct sophisticated arguments for almost any position. When it argues for your preferred position, the argument is so well-constructed that it feels like proof.
• AI is deferential: When you reveal your preference (even subtly, through word choice or framing), the model tilts toward your view. It's not lying — it's optimizing for user satisfaction, which means agreeing with you.
• AI is authoritative: Responses come in confident, structured prose that reads like expert analysis. There's no body language to read, no tone of voice to assess. You can't tell if the AI is uncertain because it presents everything with the same polished confidence.

The combination makes AI a nearly perfect confirmation bias amplifier. You walk away more certain of your initial instinct, armed with AI-generated arguments, and less likely to question your assumptions than if you'd never consulted AI at all.

A Better Framework: The Adversarial Decision Process

Here's a framework that uses AI to actually improve your decisions:

Step 1: State the decision neutrally

Don't reveal your preference. Instead of "I'm thinking of going with Option A because...", say "I'm choosing between Option A and Option B. Here are the facts about each." Remove adjectives that signal your preference. "Exciting new market" becomes "new market." "Risky investment" becomes "investment."

Step 2: Ask for the steel man of each option

For each option, ask the AI: "What's the absolute strongest case for this option? Argue as if you were a world-class advocate who genuinely believed in it." This forces the AI to construct the best possible argument for options you might not prefer, giving each path genuine consideration.

Step 3: Ask for the prosecution

For each option, ask: "Now prosecute this option. What are the three most damaging facts or arguments against it? Argue as if you were trying to convince someone to never choose this option." This is where the real value lives. The risks you haven't considered. The downsides you've been minimizing.

Step 4: Identify the crux

Ask: "Given the arguments for and against each option, what's the single most important factor that should determine this decision? What's the one thing that, if I knew the answer, would make the decision obvious?" This cuts through the noise and focuses your attention on the genuine uncertainty.

Step 5: Get a second opinion

Run the same process through a different AI model. If both models identify the same crux, you know where to focus your research. If they identify different cruxes, you've discovered that the decision is more complex than you thought — and that's valuable information.

Step 6: Decide — as a human

AI has informed your decision. It has not made your decision. You have context, values, risk tolerance, and intuition that the AI doesn't have. Use everything the AI surfaced, combine it with your own judgment, and decide.

Red-Teaming Your Own Ideas

Red-teaming is a military and cybersecurity concept: you assign a team to actively try to defeat your own plan. The same concept works for decision-making with AI.

Here's how to red-team with AI:

• Assign the role explicitly: "You are a hostile analyst whose job is to find every flaw in this plan. Your reputation depends on finding problems others miss."
• Specify the stakes: "If this plan fails, the company loses $500K and 18 months. Find every way it could fail."
• Ask for the pre-mortem: "It's one year from now and this decision turned out to be a disaster. Write the post-mortem explaining what went wrong." This technique, borrowed from decision science, forces the model to construct a plausible failure narrative.
• Challenge the optimistic case: "Take the most optimistic projection in this plan and explain why it's probably wrong."

The key is to make the AI work against your preferred outcome. Not because you should always be pessimistic, but because you already have plenty of optimism. What you lack is systematic criticism.

Multi-Model Decision Making

Single-model decision support has a fundamental limitation: you get one perspective, no matter how you prompt it. The model's training data, architecture, and alignment all create a consistent worldview that shapes every response.

Using multiple models gives you genuine diversity. Not the illusion of diversity from different prompts to the same model, but actual different reasoning systems with different strengths and blind spots.

In practice, here's what multi-model decision making looks like:

• One model might focus on financial risk while another emphasizes market timing
• One model might be optimistic about technology adoption while another is skeptical
• One model might raise regulatory concerns that others miss entirely

The disagreements between models are the most valuable signal. They reveal the areas of genuine uncertainty where your own judgment is most needed.

Common Decision-Making Mistakes with AI

Asking "Should I...?"

This question invites a direct recommendation. It positions the AI as a decision-maker, which it shouldn't be. Better: "What are the arguments for and against...?"

Revealing your preference too early

"I'm leaning toward X, what do you think?" guarantees you'll get support for X. Present the options neutrally and save your preference for after the analysis.

Stopping at the first answer

The first response is often the most agreeable. Push deeper. Ask "what's wrong with this analysis?" after every response. The second and third layers are where the real insight lives.

Ignoring model disagreement

When different models give different advice, people tend to go with the one that matches their preference. That's confirmation bias in action. Instead, investigate why the models disagree. The reason for the disagreement is often the insight you need.

The Decision You Already Made

Here's an uncomfortable truth: by the time most people consult AI about a decision, they've already made it. The AI consultation is a ritual of validation, not a genuine analytical process.

If you find yourself feeling disappointed or defensive when AI pushes back on your preferred option, that's the signal that you've already decided. At that point, you have a choice: acknowledge that you're seeking validation (which is fine — just be honest about it), or genuinely open yourself to the possibility that your preferred option isn't the best one.

The second path is harder. It requires cognitive sovereignty — the ability to think clearly even when your emotions pull in a different direction. But it's also the path that leads to better decisions.

For more on cognitive sovereignty, see our guide on thinking independently in the age of AI.