When USA TODAY asked Microsoft Copilot to simulate the rest of the women’s NCAA tournament a week ago, the AI confidently predicted all four No. 1 seeds would reach Phoenix. With that call now validated, the updated April 2 forecast gives UConn a clear edge to win it all—but also shows a Texas team that’s far more than just a runner-up in the model’s eyes.
What the Simulations Actually Show
Microsoft Copilot first ran its bracket simulation on March 26, when the Sweet 16 was set. The AI correctly projected that South Carolina, UCLA, Texas, and UConn would emerge as the Final Four. A week later, with the field reduced to those four top seeds, USA TODAY asked for an update. The results, published April 2, hold UConn as the most likely national champion—but the details reveal a more layered story.
In the semifinal between UConn and South Carolina, Copilot has the Huskies advancing in roughly two-thirds of its simulated runs. One sample scoreline had UConn winning 78–68. The other semifinal pits Texas against UCLA, and here the model sees something closer to a coin flip: Texas moves on a bit more often, but the margin is slim. A sample simulation gave the Longhorns a 74–71 win.
The championship projection then pits UConn against Texas, with the Huskies winning a clear majority of the time. A sample final had UConn on top 80–70. But here’s a twist: when USA TODAY asked Copilot for a single random prediction—not a probability distribution—the AI flipped the script and had Texas beating South Carolina 68–63 for the title. That inconsistency isn’t a glitch; it underscores how even a model that leans heavily toward one favorite can produce a different result on any given run.
Key Prediction Details
| Matchup | AI Edge | Sample Score |
|---|---|---|
| UConn vs. South Carolina | UConn in ~67% of runs | UConn 78–68 |
| Texas vs. UCLA | Texas slight favorite | Texas 74–71 |
| Championship (UConn vs. Texas) | UConn clear majority | UConn 80–70 |
What This Means for Your Final Four Viewing
If you’re tuning into the ESPN doubleheader on April 3, the AI predictions give you a practical lens for what to watch. They’re not a crystal ball, but they highlight which matchups are likely to turn on small margins and which teams have the statistical profile to control the game.
For casual fans: The UConn–South Carolina semifinal is the heavyweight bout. The AI’s two-thirds edge for UConn is meaningful, but it’s not a lock. If you’re looking for the most likely single-game upset, this is the one. A South Carolina victory would be a clear departure from the model’s central expectation, but the Gamecocks have the depth and coaching to pull it off. Watch for how UConn handles South Carolina’s physicality early—if the Huskies build a lead, they’re built to protect it.
The Texas–UCLA game is a toss-up. The model treats this as the more volatile semifinal, which means you should expect a game that might be decided by a single hot streak or a critical officiating call. Neither team is a pushover; both have the talent to win it all. If you’re looking for a nail-biter, this is the game.
For bracket-challenge players: If you’re in a pool that’s still alive, the AI intel suggests UConn is the safest champion pick. But if you need to differentiate from a crowd that’s likely heavy on the Huskies, Texas as a champion gamble has a plausible path—and the model says that path is more realistic than a deep run from UCLA or South Carolina. Just know you’re betting against the numbers.
For bettors (where legal): The simulation percentages aren’t real-time betting odds, but they can help you spot value. If the market implies UConn is an even bigger favorite than Copilot’s ~67% in the semifinal, there may be room for a South Carolina wager. Conversely, if Texas–UCLA is priced near even, the model’s slight lean toward the Longhorns might reinforce that it’s basically a fair fight.
How We Got Here: AI and March Madness
Microsoft has been integrating Copilot into sports for a while. The tool can surface live scores, summarize games, and—when prodded—run simulations based on current data. The USA TODAY experiment taps into that capability, but it’s worth understanding what the model is actually doing.
Copilot’s simulation is not a complex, physics-based replay of every possession. It’s a probability engine that weighs factors like team efficiency, recent performance, and likely matchup dynamics. The more data it has, the more reliable the output—which is why a Final Four of all No. 1 seeds is actually a better scenario for the AI. There’s a rich history of how these programs perform at the highest level, giving the model plenty of signal to process.
The AI’s preference for UConn isn’t a mystery. The Huskies combine elite talent, a tournament-tested roster, and a statistical profile that tends to produce stable outcomes in simulation models. They limit turnovers, score efficiently, and defend at a level that makes them less susceptible to the wild swings that can doom other top seeds. In a short tournament, that stability is valuable, and the model rewards it.
But as the Texas–UCLA coin flip shows, the AI also acknowledges when two teams are nearly indistinguishable on paper. That’s an honest reading, not a failure of analysis. The model is saying that this game could easily go either way, which is exactly what makes the Final Four compelling.
What to Do Now: Watch Smart, Not Superstitious
With the games just hours away, here’s your service-journalism playbook:
- Take the probabilities seriously, but not literally. UConn advancing two-thirds of the time means there’s a real one-in-three chance they lose. Don’t be shocked if South Carolina pulls it off.
- Use the AI as a scouting report. The model says UConn is more likely to win because of its balance and crunch-time execution. When you watch, pay attention to how the Huskies handle the last five minutes of a close game. If South Carolina can force them into rushed possessions, the odds shift.
- Enjoy the uncertainty in the Texas–UCLA game. Embrace the fact that this is the ultimate toss-up. The AI’s slight edge for Texas might be based on something as subtle as defensive rebounding rates. Look for which team controls the glass and gets second-chance points—that could be the difference.
- Ignore the random-prediction drama. The fact that Copilot spit out a Texas-over-South Carolina title when asked for a blind pick is a reminder that simulations produce a range of outcomes. Don’t read too much into it; it’s not the model “changing its mind.” It’s just one roll of the dice.
What to Watch Next
The immediate answer is the games themselves. On April 3, the two semifinals will either validate or embarrass the AI’s read. If UConn and Texas both win, the model’s core narrative holds. If one or both underdogs break through, it’s a lesson in how even a well-fed model can’t capture everything that happens in 40 minutes of basketball.
Beyond this weekend, watch for how sports media—and Microsoft—continue to use AI predictions. The technology is clearly improving, and these simulated brackets are becoming a staple of tournament coverage. The next frontier is live, in-game modeling that updates probabilities possession by possession. For now, Copilot’s Final Four forecast is a useful, if imperfect, conversation starter. The real champion will be decided not by code, but by players making plays in Phoenix.