蘑菇视频网

One of the coolest things about NCAA Tournament bracket challenges is you don鈥檛 need to 鈥渒now ball鈥� to win. With 68 teams in the field and 67 total games to predict, incredible upsets 鈥� when a lower-seeded team beats a higher-seeded team 鈥� are expected, and the most dedicated college hoops fan鈥檚 bracket can be foiled within the first few hours of the tournament as a result.

According to 蘑菇视频网 statistical analysis lecturer Chris O鈥橞yrne, who first explained to SDSU NewsCenter in 2024 the actual odds of predicting all 67 games correctly come out to 1 in 147,573,952,589,676,412,928 (2 to the 67th power), or a few quadrillion more than 1 in 147.5 quintillion. If you鈥檙e skipping the tournament鈥檚 鈥渇irst four鈥� games, the odds stumble to 1 in 9,223,372,036,854,775,808, or a little more than 9.2 quintillion (2 to the 63rd power).

With considerable advances in and adoption of AI since 2024, many may turn to platforms like ChatGPT for an edge in predicting games, or even for help submitting extra brackets. O鈥橞yrne helped us put these theories to the test. He prompted four different AI platforms to fill out brackets of their own, and came up with some mathematical and some practical reasons why AI isn鈥檛 likely to get you any closer to predicting a perfect bracket.

Can AI help me submit thousands, millions or billions of different bracket variations?

鈥淏etween ChatGPT, Copilot, Claude and Perplexity, not one of these platforms was able to fill out a copy of a blank bracket provided by the NCAA,鈥� O鈥橞yrne explained. 鈥淚 ended up asking the platforms for their predictions in list form. Based on this experience, it would be a tall order to ask AI to complete a dozen brackets, let alone 100 promising ones.鈥�

Let鈥檚 say you found a platform that could mass-produce a selection of brackets. Where would you start? Three of the four AI platforms O鈥橞yrne selected picked all four regional No. 1 seeds to advance to the Final Four. If you followed the same strategy, you鈥檇 still have 43 more games to predict correctly.

鈥淭hose odds of doing that shrink down to 1 in 8,796,093,022,208 (nearly 8.8 trillion),鈥� O鈥橞yrne said. 鈥淵ou can reduce the number of possibilities by making reasonable assumptions 鈥� like advancing top seeds 鈥� but you can鈥檛 eliminate the uncertainty.鈥� 

, the four regional No. 1 seeds in a given tournament have all advanced to the Final Four only twice, in 2008 and 2025, since the tournament field expanded to 64 teams in 1985.

If AI can scour the internet for expert analysis, it can make 鈥渟marter鈥� picks than I can, right?

Theoretically, yes, O鈥橞yrne said. But it鈥檚 still not enough to make a difference.

鈥淭he reality is, expert or not, there鈥檚 a greater than 99% chance you won鈥檛 have a perfect bracket after the first round,鈥� O鈥橞yrne said. 鈥淧robably closer to 99.5%.鈥�

O鈥橞yrne explained this point with another simple math problem. The first 24 hours of the tournament feature 16 of the 32 first-round games, which, according to O鈥橞yrne鈥檚 math, means there are 65,536 different possibilities (2 to the 16th power) for those games alone. 

鈥淚f you treated every game like a coin flip, just getting the first 16 games right would be about a one-in-65,000 chance,鈥� O鈥橞yrne explained. 鈥淎nd that鈥檚 assuming every game is 50-50 鈥� which they鈥檙e not. Even under ideal conditions, the number of possibilities is enormous.鈥�

Even if you assume the pair of No. 1 seeds and the pair of No. 2 seeds that play on day one will beat their far inferior opponents, you still only have 1 in 4,096 (2 to the 12th power) odds of predicting the other 12 games correctly.

鈥淎fter just the first day of games, typically less than 0.1% of brackets remain perfect,鈥� O鈥橞yrne said.

Read SDSU NewsCenter鈥檚 previous articles on March Madness odds:

March Madness I: A statistician鈥檚 guide for beating 1-in-147-quintillion odds of the perfect bracket (2024)

March Madness I: Breaking down odds of predicting the perfect bracket (2025)