I'm not a college basketball fan, but when the NCAA tournament rolls around — aka March Madness — my interest grows, mostly around making the celebrated Bracket.
In the US, it seems like everyone is in a pool, filling out brackets in the hope that they can guess the winners and losers and walk away with bragging rights or a pool prize (usually monetary); Apple Sports has also got in on the bracket action. How people cook up their brackets varies widely. Some follow college sports closely enough to have a good idea of which team(s) might triumph. I am not one of those people — I'm not even a casual college hoops fan.
It's reasonable to assume these days that some people are using AI to build their brackets, and I wondered if I could do the same. With the tournament kicking off on March 19 (and running through April 6), I was too late to build one and submit it to a pool; but as an exercise, I could see if AI game, rounds, and championship selections were any better than my dead reckoning.
I built a prompt and fed it into ChatGPT Plus and Gemini Pro. In each case, I chose the 'Thinking' mode and, as you can see, I encouraged the AI to take its time.
Here's the prompt:
"I'd love to see the optimal NCAA tournament bracket, one that has the best chance of coming true. Don't rush on this. Consider records (season, historical), external factors (any you think are relevant). I want to see the whole bracket and, if possible, be able to click on a game or round to see your reasoning."
I started with ChatGPT. For more than five minutes I watched as it worked through the problem, including deciding where it could source a decent 2026 NCAA men's tournament bracket in PDF format, which it used to extract the text.
Next, it looked at current tournament odds, favorites, and stats from sites like Reuters and Maizenbrew. It looked at seed data from The NY Post and The Telegraph (among others). ChatGPT found tournament performance details and information on coaches from the NCAA and Bracketodds.cs.illinois.edu.
It spent a chunk of time looking at injuries.
Next, ChatGPT used the data to make what it termed "reasonable tournament picks," but also considered historical upset data. More than once, it mentioned something called "chalk" or "chalk leaning", which in betting parlance means "favorites" or favored to win. ChatGPT was chalky throughout.
I could see that it started favoring some teams over the others: early favorites were Duke and Purdue. It kept a smattering of upsets in its plan, but definitely favored the safe, rational (chalky) approach over any Cinderella story.
Finally, ChatGPT was ready to build the bracket and, on its own, chose to make a web page, using Python, HTML, and some simple CSS.
Image
1
of
3
Next thing you know, I had an HTML file that I could download and plug into Safari. It's a stunning piece of work. Each matchup showed a winner, and I could scroll down through regions and rounds. For each game and pick, I could click to expand and see ChatGPT's reasoning.
In a matchup between Virginia and Iowa State in which Iowa wins, ChatGPT wrote, "Iowa State has the stronger Final Four signal and feels more explosive while remaining sturdy enough defensively." Granted, I don't know enough about college basketball to know if this is accurate.
Near the bottom of the page, I found that the final matchup would be between Arizona and Michigan, with the former taking the NCAA Championship. ChatGPT explained its reasoning, but also told me which upsets could destroy the bracket:
"Highest-confidence regions: East early chalk, Midwest top line.
Swing games that could break this bracket fastest: Wisconsin/High Point, North Carolina/VCU, Tennessee/Miami (Ohio), Kentucky/Santa Clara, Duke/Arizona.
Different AI, same result
Gemini Pro was quicker, returning a more skeletal table that I could export to Google Sheets and then its 'In-Depth Reasoning'. It took into account my hometown and why I might keep an eye on St. Johns, but also shared why Duke would survive to the Final 4.
It predicted some upsets, like Santa Clara taking down Kentucky, but in the end, Gemini reached the same conclusion as ChatGPT: Arizona would win it all:
"This is a "Heavyweight" final. Duke has the star power (Boozer), but Arizona has the roster versatility. Arizona's ability to throw three different elite defenders at Cameron Boozer will be the difference.
The Prediction: Arizona wins its first title since 1997 in a high-scoring, 82-76 classic."
I wasn't that impressed with Gemini's work, but I loved what ChatGPT built (sadly, I cannot load the page in here, but you can see screenshots above). So, I asked ChatGPT to keep an eye on the tournament and let me know how the bracket is holding up. I wanted ChatGPT to watch Arizona. If it lost in an upset in one of the early rounds, this bracket would be a bust.
"Done — I’ll check it every evening through the championship game and flag Arizona immediately if they go down. The tournament runs through Monday, April 6, 2026."
There's always an enormous element of chance in a tournament like this, but I'm curious to see if ChatGPT and Gemini's picks are any better than mine or yours.
Let me know your March Madness thoughts in the comments below.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course, you can also follow TechRadar on YouTube and TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
