March Madness AI Bracket Showdown: ChatGPT vs. Humans

March Madness AI Bracket Predictions: ChatGPT's Final Four Picks

The AI revolution is here with seemingly no end in sight - but can artificial intelligence pick the perfect March Madness bracket? We are once again turning to ChatGPT for its March Madness AI bracket predictions and projected final score of every game in the NCAA Tournament. Last year, we trained the AI model to predict the entire bracket and final score for every game in the tournament. ChatGPT rewarded us by correctly picking the national champion in UConn, two of the Final Four teams, and 21 of 32 first-round games - including three of its six upset picks involving No. 11 seeds or lower. Since then, OpenAI has upgraded ChatGPT by leaps and bounds. Could the latest AI model finally nail that elusive perfect bracket? Or do humans still have the edge for three magical weeks in March? Here's a look at our AI bracket predictions with exact score predictions for all 63 matchups in the 2025 NCAA Tournament.

Training the AI Model

If you're looking to replicate this project at home, it's entirely possible to have ChatGPT fill out your March Madness bracket. It just takes a little work upfront. First and foremost, I would recommend using GPT-4o, which is a more advanced model of ChatGPT that relies on real-time search data to fill in the gaps of its preexisting knowledge base. It's also critical to train the large language model - which knows data, not basketball - with anything that might be relevant to making March Madness predictions.

Before we asked ChatGPT for its March Madness bracket predictions, we established a set of basic criteria and methodology to instruct this project. This is always an important step when using a chatbot, which is designed to follow instructions without necessarily understanding what it might be asked to do in the first place. In this case, we asked ChatGPT to predict every game of the NCAA Tournament using the same factors that others consider when making their college basketball picks. We asked it to make these predictions based on team-specific and matchup data as well as the latest college basketball odds and historical trends - all of which served as the basic training data for this bracket projection.

Critically, we also trained ChatGPT on the most common March Madness upsets to reflect the typically chaotic nature of the NCAA Tournament. It wouldn't be a March Madness bracket if all of the favorites won, right? To be clear, this is for entertainment purposes only, as AI chatbots are not a replacement for typical sports betting models or even the basic research that many do when filling out their bracket. That said, ChatGPT can excel at spotting trends or other data points that humans might overlook, so use these AI picks at your own discretion!

Feeding Data to the AI Model

With the guidelines for this project established, we turned to feeding ChatGPT as much training data as possible to improve the "accuracy" of its picks. First, we asked it to search for common trends related to the NCAA Tournament while also sharing with it some of the best March Madness bracket advice from our expert staff. Then, we provided it with a full list of March Madness odds and Final Four odds for all 68 teams in the tournament field as of Selection Sunday.

Additionally, we imported a spreadsheet of KenPom data for every team in the 2025 NCAA Tournament, including how each team ranks in tempo, adjusted offensive and defensive efficiency, and adjusted efficiency margin. This involved a few rounds of training to ensure accurate recognition and interpretation of the data, which is a crucial step when working with any developing model.

Finally, we provided ChatGPT with a list of updated spreads and totals for all 32 first-round games to help establish a baseline expectation from the betting market. This year, we also utilized ChatGPT's brand-new "Deep Research" function to allow it to further research NCAA Tournament stats and trends before locking in its picks.

Then, we finally let the robot loose to predict this year's bracket. And boy, did it deliver.

Upsets and Predictions

If you were turning to artificial intelligence in hopes of escaping the harsh reality of the real world, I have some unfortunate news: the robot is picking Duke, too. Duke features three of the best players in March Madness led by star freshman Cooper Flagg, who makes an appearance more than a few times in our AI predictions. But it wasn't all chalk for ChatGPT, which only picked two No. 1 seeds in the Final Four and predicted seven double-digit seeds to pull off first-round upsets.

What would an NCAA Tournament bracket be without a few upsets? Fortunately, ChatGPT had the same idea, picking 13 lower-seeded teams to advance in the bracket - including nine in the first round alone. Here's a full list of every upset ChatGPT predicted in the first round, ranked by seed differential:

Based on the latest betting odds, KenPom data, and other March Madness trends, here are ChatGPT's picks and score predictions for every game of the 2025 NCAA Tournament.

Auburn's relentless pressure and transition offense prove to be the difference in a tight battle with Texas Tech. The Red Raiders keep it close with their gritty defense and timely 3-point shooting, but Auburn's ability to force turnovers and control the tempo ultimately prevails. Late-game free throws seal the Tigers' spot in the national championship game.

Prediction: Auburn 75, Texas Tech 71

Tennessee's physical defense slows Duke's high-powered attack for much of the game, but Cooper Flagg and the Blue Devils find answers in key moments. Duke’s superior shot-making and ball movement allow it to break down Tennessee’s half-court defense, while its own defensive adjustments in the second half stifle the Vols’ scoring threats. Clutch buckets down the stretch send Duke to the national title game.

Prediction: Duke 74, Tennessee 68

In a thrilling national championship showdown, Duke’s offensive firepower and defensive adjustments prove just enough to edge out Auburn. The Tigers push the tempo early, using their depth and defensive pressure to create turnovers and easy buckets in transition. However, Duke weathers the storm behind the veteran leadership of Tyrese Proctor and Sion James and the all-around brilliance of freshman sensation Cooper Flagg.