Prompt AI to be your biggest critic

May 4, 2026

"If I had a magic wand and could eliminate one bias, it would be overconfidence."

~ Daniel Kahneman (1934–2024)

I asked AI to tell me why I was wrong about a stock I liked. It did.

Most investors use AI to efficiently gather information in support of a thesis. The better investors use it to blow holes in it.

I now use AI as my designated bear. I explicitly ask it for the strongest arguments against my own ideas.

Some asset managers are already moving in this direction, using AI not just to gather data but to challenge their own decisions. The edge is not more information. It is better decision-making under pressure.

Man Group, the $160B+ global quant manager has designed what they call, AlphaGPT. It takes the form of a three-agent team to develop and test a new investment idea. The three AI personalities are “The Idea Person,” “The Implementor,” and “The Evaluator.” The idea person develops an investment thesis. The implementor writes the code to test the idea. The evaluator plays the role of skeptic to the work of the other two. The goal is not just to find ideas, but to challenge them.

I’ve started applying a simpler version of this process in my own work. Several months ago, I was researching Nike (NKE), a stock I liked at the time. I gave ChatGPT my thesis and made a simple request: “Please give me your best arguments on why I am wrong.” A few seconds later, I got this response:

1. Structural Decline in China Sales.

2. Brand Dilution and Competitive Pressures.

3. Innovation Gap.

4. Margin Pressure from DTC Push.

5. Tariff Risk Not Fully Discounted.

6. Valuation Still Not Cheap.

7. Management Concerns and Execution Missteps.

Each of these topics came with three supporting bullet points. While I regarded several of these arguments as well-documented and reflected in the price, they couldn’t all be dismissed out of hand. I cut part of my position. The shares have continued to be weak since, down 30% year to date.

To get the most effective responses, I recommend users of AI chatbots go into their account settings and give customized, standing instructions for how they wish to receive information. I personally adopted the framework suggested by a friend who is the head of a private equity firm and uses AI in his due diligence process. These include:

• Always provide the pros and cons of something if you can. Be critical.

• Suggest solutions that I didn’t think about. Be proactive and anticipate my needs.

• Be opinionated rather than neutral when appropriate.

• Value good arguments over authorities; the source is irrelevant.

• Consider new technologies and contrarian ideas, not just conventional wisdom.

• You may use high levels of speculation or prediction, just flag it for me.

• After a response, provide five follow-up questions formatted as bullet points.

The late Daniel Kahneman regarded overconfidence as the most damaging bias since it compounds the effects of all other biases and is resistant to learning and course corrections. AI is uniquely suited to combating overconfidence because it has no ego, no career risk, and no social relationship with you. The key is prompt construction - a poorly worded prompt will get validation, not criticism. “Please give me your best arguments on why I am wrong” is specifically designed to override the natural tendency of AI to affirm the user’s prior view.

“Prompt AI to be your biggest critic,” is one of 69 solutions discussed in my forthcoming book, Running Against the Herd: Battling Biases to Make Better Investment Decisions, now available for pre-order.