If you spend much time scouring the internet for sports betting content, you’ll invariably find many articles showcasing AI picks. It’s a somewhat grueling exercise, as much of the content is click-bait (at best) or (at worst) pure junk.
One thing you’ll note, however, is that many AI predictions, particularly those made at the beginning of a season, tend to be unimaginative, broadly following the predictions you will find on mainstream sports platforms and the sports betting odds in general.
For example, looking back at early October 2024, you will see many articles promoting AI’s picks for the season. Most of them pointed to a Boston Celtics vs OKC NBA Finals. As we know, they were half right. But that’s hardly a coup, as those two teams were backed mainly as the favorites in the Eastern and Western Conferences.
NHL season went by the formbook
In the NHL, which does get less AI treatment compared to basketball, soccer, and football, many AI pieces did ‘predict’ that the Florida Panthers and Edmonton Oilers would meet in the Stanley Cup Final. But again, this shouldn’t be cause for a victory lap, as those two teams – last year’s finalists – were the preseason favorites in the 2025 Stanley Cup odds, so it’s not a great leap of faith.
The point, as such, is that AI is not departing from the usual content you could read on a betting blog or mainstream sports broadcaster’s website. It would be impressive if an AI prediction said that the Indiana Pacers would defy the odds and go to the NBA Finals, but you will be hard-pressed to find any.
In a sense, it shows the limitations of AI. When you ask ChatGPT or Gemini to predict the 2025/26 NFL season, what do you think the AI does? It will look at a mix of betting content and analysis from major sports media outlets (those that it has permission to view) and come up with an answer. You can be sure its reasoning won’t differ much from what you see in the betting odds.
How to get AI to build a worthwhile response
Of course, none of this should be surprising. Sportsbooks don’t pull odds out of thin air, and the logic and algorithm-based data that goes into the actuary of compiling those odds is fundamentally the same in terms of metrics that an AI will look at. In short, if AI has access to the same data – a mix of complex data and opinion – that the rest of us have, it is unlikely to have a eureka moment.
The question, then, is how do we get to a point where AI can tell you that the Pistons will have a surprisingly good season when nobody else – or at least very few – is saying the same thing. The short answer is that we can’t get there, as that is how sports work. However, there is a longer answer that is full of caveats and much more optimistic about the future of AI and sports predictions.
The first is that, as mentioned, AI is only as good as the data it has access to, so it is possible to feed it good data from various sources, such as subscription-based sports betting data services. If you can get access, you could even feed it the raw data that sportsbooks use to compile odds. The LLM (large language model) that underpins an AI chatbot can sift through this information, potentially millions of data points, arriving at some more interesting conclusions than regurgitating content from an ESPN article.
Secondly, and most importantly, AI depends on the prompts’ quality. So, if you are asking, “Who will win the Super Bowl next season?” you are probably going about things the wrong way. If you ask it to look at the betting odds for each team, compare it to the data sets you have received, and pinpoint value in the odds, that’s the point where the AI may cite a team like the Pistons to defy the odds.
The thing is, all of this takes planning and hard work, as well as the cost incurred to buy professional data. Yet, it is a logically sound approach. It is much more complex than reading articles titled, “We asked ChatGPT to name the next 10 NBA Championship winners,” but it could be much more rewarding.
