How to Stress Test Your Opinions With AI

Most people use AI to confirm what they already believe, which is exactly the wrong move. The better play is to use it to dismantle your own position, rebuild the opposing one from scratch, and then watch the two collide. Whatever survives that collision is worth keeping. The rest was thinner than you thought it was.

Here's the method I actually use whenever I want to pressure-test a strong opinion.

The three-chat method

You need three separate conversations — not one, three. The separation is the whole point.

Chat 1: build the strongest case for Side A. Tell the AI it's prepping for a debate, it's arguing for Side A, and its job is to win. Have it research the best arguments, anticipate counterattacks, and put together the most compelling version of that position it can. No softening. You want the steel man, not the straw man.

Chat 2: build the strongest case for Side B. Same setup, opposite direction. Best arguments, best counterarguments, best research. Let it swing as hard as it did for Side A.

Chat 3: stage the debate. Drop both documents into a fresh conversation. Have the AI run the debate, then step back and judge it: where does Side A win, where does Side B win, where are both sides actually weak, and where do they secretly agree?

The third chat is where the real learning happens.

Why three chats and not one

Ask a single chat to "give me both sides" and you'll get a lukewarm list. The model hedges from the jump, tries to stay balanced before either side has even spoken, and neither position gets its best shot.

When you instead force the model to commit fully to one side, it digs in. It finds arguments you've never seen, anticipates objections you didn't know existed, and builds a case with actual teeth. Do the same on the other side and you end up with two documents that are each genuinely strong — and the debate between them is real instead of performative.

What happened the first time I tried it

I ran this on a political topic where I had a firm, well-defended opinion.

The AI built a case for my side that was, honestly, better than anything I could have written myself. It pulled data I'd never seen, framed arguments more cleanly than I had, and anticipated counterpoints I'd never thought to plan for. Then it did the same thing for the other side — and that second document was uncomfortable. Not because it was wrong. Because it was good. There were arguments in there I had no answer for, points I'd never seriously engaged with, and data that genuinely complicated my position.

When the third chat compared the two, the verdict wasn't "your side wins" or "their side wins." It was more like: Side A is stronger on these three issues, Side B is stronger on these two, and on this one they're both making the same mistake from different angles. I'd never read an opinion piece anywhere that left me with a cleaner picture of a topic I cared about.

What this is actually good for

Politics is the obvious use case, but it's not the most useful one. You can run this on almost anything where you hold a position:

• a business decision you're leaning toward
• a disagreement with a partner or coworker
• a strategy choice between two approaches
• an argument you keep having with the same person
• a belief you've never seriously questioned

The shape is always the same: force the strongest version of each side into existence on its own, then let them fight on equal footing.

The fourth move

After the debate, there's one more step worth running. Ask the AI:

"Now that you understand both positions fully, what would you actually recommend? What's the most honest, well-informed path forward?"

This is where synthesis happens. The model isn't advocating for either side anymore — it has the full picture, and its recommendation usually pulls the strongest pieces of both positions while quietly dropping the weak ones. It's not gospel. It's just a much better starting point than whatever you walked in with.

What it did to my thinking

Two things happened that I didn't expect.

I got more humble. On most topics I'm confident about, I really only know two or three points well — usually the ones backed up by my own stories and experience. There are a dozen other points I've never seriously engaged with. The stuff I was sure about was real, but it was a small piece of a much bigger picture.

I got more curious. Once you've seen how much stronger your thinking gets when you actually wrestle with the other side, you start wanting to do it more. The reflex shifts from "how do I defend my position" to "what am I missing here."

That second shift turned out to be worth more than winning any argument I've ever had.

The practical takeaway

You don't have to be neutral on everything. Strong opinions are fine — necessary, even. But strong opinions held without stress-testing are just habits in formal wear.

AI lets you pressure-test any belief in about thirty minutes. No debate partner, no ten-book reading list, no waiting for the right argument with the right person. Just three chats and the willingness to let the other side make its best case. The opinions that come out the other end are the ones worth keeping. The ones that fall apart were never really yours to begin with.