We are reaching the conclusion of the 2025-2026 NFL Season, with Super Bowl LX on the horizon. The big game will take place on February 8 at Levi’s Stadium. While we will have to wait for the conclusion of the game, as well as the Playoffs, to be able to see just how good – or bad – the season’s predictions have been, it’s fairly evident that AI models had a difficult time of it. A lot of AI predictions were published for the NFL season, and, by and large, the predictions were just as bad, at times worse, than the human experts’ predictions.
To be fair, this has been a strange NFL season. Arguably, it has been one of the best ever campaigns, especially if you like shocks, underdog victories and stirring comebacks. Yet, the upshot of that is that the season made a mockery of expert predictions made before the campaign began in September. Of course, it is normal to see teams’ fortunes wax and wane as a season progresses, but this season has had more surprises than normal, and a lot of big favorites struggled.
Super Bowl betting odds varied wildly across the season
A case in point can be seen in tracking the Super Bowl betting odds from DraftKings from the opening week to the eve of the Super Bowl. At the beginning of the campaign, it was almost a four-way tie for the favorite spot between the Baltimore Ravens, Buffalo Bills, Kansas City Chiefs and Philadelphia Eagles. That most certainly did not go to plan. The Ravens and Chiefs did not even make it to the Playoffs, and both the Eagles and Bills have been eliminated from contention. Other teams high up in the odds preseason, like the Detroit Lions, Green Bay Packers and Washington Commanders, were also eliminated early.
One of the interesting aspects to all this is that AI broadly mimicked the betting odds when laying out predictions. It gives an insight into how most (but not all) AI models are being used to make predictions, i.e., by effectively looking at the human data and analysis from sportsbooks and sports websites. In short, asking AI to make a prediction is basically like asking it to find a consensus from all the freely available data and expert opinions available on the internet.
Better data would be needed for AI to get ahead
In saying that, it is most certainly possible that AI models can use proprietary data to make predictions, gaining access to specialized data sets that may have told us that teams like Seattle Seahawks, LA Rams, New England Patriots and Denver Broncos would shine, whereas the preseason favorites would fall by the wayside. But, once again, this is the sort of specialist data that is used by sportsbooks, who might also use AI to formulate odds, so it’s not fully evident that specialist AI models would have fared better.
<iframe width=”560″ height=”315″ src=”https://www.youtube.com/embed/vP-lB6d-V3I?si=3tp0Lr7UfylqDsSR” title=”YouTube video player” frameborder=”0″ allow=”accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share” referrerpolicy=”strict-origin-when-cross-origin” allowfullscreen></iframe>
In the end, it’s immediately apparent that consumer-facing AI is not ready to compete in the arena of sports predictions. That doesn’t mean the models are useless. You could, for instance, ask it to run certain calculations, showing you how specific bets offer more value than others. Yet, we are a long way off any sense of clairvoyance on AI’s part. This crazy NFL season told us that when it comes to sports predictions, AI is just as fallible as we humans are.
