ChatGPT’s Boring Bracket Probably Won Your March Madness Pool

When it comes to winning an NCAA Tournament bracket pool, some luck is generally involved if you want to win.

Or maybe not. 

If you used AI to fill out a bracket this year, you probably would be collecting cash today, at least that is the takeaway from analyzing the bracket ChatGPT filled out for VegasInsider.com. 

Michigan wins it all, as predicted by AI

If it had been in the ESPN Tournament Challenge, ChatGPT would have finished in the 99th percentile when all was said and done. 

That is primarily for two reasons: Michigan won it all, and there were not many upsets along the way. 

Going into the final weekend, that particular ChatGPT-influenced bracket would have been in the 92nd percentile with 910 points, but a lot was riding on the Wolverines finishing the job because a massive amount of points ride on the last few games. 

Michigan’s win over Arizona in the national semifinals was worth 160 points, and the Wolverines’ victory over Connecticut on Monday night brought in another 320 based on the ESPN scoring system, which is widely used in bracket competitions. 

That pushed the AI bracket up the leaderboards because a plurality of users picked Duke to win it all, and Arizona was also a more popular pick among the public at ESPN. 

Only 14.19% of ESPN users picked Michigan while 23.1% picked Duke and another 20.85% picked Arizona. 

So the ChatGPT bracket finished with 1390 points while the winner of the ESPN Bracket Challenge had 1810 and five more people finished with 1800. 

Also of note: The AI bracket would have beaten everyone on the celebrity board, edging ESPN anchor Matt Barrie’s bracket by 70 points. 

What could have been? 

The ChatGPT bracket would have done even better if it had accurately predicted Florida’s early demise at the hands of Iowa in the second round, but that is more or less collateral damage considering overall the bot clearly was going by the numbers that logically dictate barely any upsets will occur.  

(That's why they are upsets, of course.)

As some human experts predicted, this turned out to be a fairly chalky tournament, so it was better to be AI relying on the cold, hard data than a human who just can’t help trying to guess which Cinderella might survive past midnight (or at least the first round). 

Ultimately AI had two Final Four teams — Michigan and Arizona — and four of the Elite Eight. 

Illinois upsetting Houston in the Sweet 16 was also significant as the ChatGPT bracket included all four No. 1 seeds, three No. 2s and a No. 3 (Gonzaga). 

More on the process 

According to a press release, VegasInsider fed ChatGPT various datapoints including win-loss records, seeding, offensive and defensive performance, conference strength, player production, coaching experience and recent tournament history. 

In all, over 1,700 datapoints were used, and they say the bot produced 71 pages and 10,000 words to justify the picks.