Why Four AI Models Picked the Same Winner — and Couldn't Agree on Anything Else

The Sweet 16 tips off tonight. Your bracket is probably already busted.

But here's what most people don't realize: you can hand the exact same question to four different AI models and get four meaningfully different answers — not because they're wrong, but because they think differently.

The Wall Street Journal just proved it. They secretly entered Claude, ChatGPT, and Gemini into their office bracket pool. Same data, same rules. Claude went contrarian and picked No. 3 Illinois to win it all. ChatGPT and Gemini played it safe with No. 1 Michigan. After the first weekend, Claude is ranked 6th out of 124 entries — beating more than half the humans.

TLDR: We gave Claude, ChatGPT, Perplexity, and Grok the same March Madness prompt and compared their answers side by side. All four picked Houston — but their confidence, risk tolerance, and reasoning styles were wildly different. The real insight: AI models have personalities, and knowing which one to use for what is the new power skill.

So we ran our own experiment.

We built one prompt and ran it through four leading AI models — asking each to pick tonight's No. 2 Houston vs. No. 3 Illinois Sweet 16 game. But here's the twist: we didn't just ask for a pick. We asked each model to name itself, rate its own confidence, describe its reasoning style, admit what it might be wrong about, and rate its own risk tolerance on a 1–10 scale.

Most people don't know you can ask AI to do any of this. You can. And the answers are fascinating.

What This Actually Tells You

All four picked Houston. Same winner. But that's where the agreement ends.

Perplexity was the most confident at 72% — pulling live odds and betting lines to back it up. Grok was the least confident at 58% despite calling itself "ruthlessly analytical." And Claude? It rated itself a 3 out of 10 on risk tolerance while every other model said 6. When Claude says "I'd rather be right than clever," it means it.

The biggest reveal: all four almost picked Illinois and talked themselves out of it. That's not a coincidence — it tells you the models are genuinely wrestling with uncertainty, not just spitting out a default answer.

And three out of four would bet their reputation on Purdue over Texas tonight. ChatGPT went rogue and picked UConn instead. Even AI models have hot takes.

Want to run your own AI model showdown? Use this on Claude, ChatGPT, Perplexity, or any AI:

The trick isn't the basketball — it's the questions most people never think to ask. Asking an AI to rate its own confidence, admit what it might be wrong about, and describe its reasoning style turns a generic answer into a window into how each model actually thinks.