Is your fantasy football team floundering like a floppy fish out of water? Are you trying to salvage your season after drafting some duds? Or are you just trying to get a leg up on the competition this year to avoid humiliation come December?
AI can probably help you with that. But to what extent—and how far do you want to take it? Although using AI for sports analytics isn’t a new concept, it’s certainly been heating up as the overall appetite for AI has grown exponentially since ChatGPT was born.
Let’s take a look at how to use AI for fantasy football, including some real results (and opinions) from some people who have put it to the test. (Side note: We’re an AI analytics company for enterprise companies, not for fantasy footballers. We just think it’s interesting to read about. 😄 )
In This Post
The history of AI for sports analytics
If you tune in to any professional sports game on TV, there’s probably some sort of AI technology happening behind the scenes to gather stats. If you watch baseball nowadays, you’ll see Statcast hard at work, which uses Google Cloud AI to track useful stats like velocity rates of hits, bat speeds, pitch speeds, and more, benefiting fans, players, clubs, and data enthusiasts alike. It even compiles some of the most obscure data you could imagine, like how hot dog sales can show how a team is doing. (Going to cite Google Cloud on this one—apparently, if the home team is playing well, fans will buy more ‘dogs.)
For sports analytics and fantasy football in particular, predictive models have always been involved, but early systems (we’re talking pre-2010) relied more on statistical approaches like regression models or simple, heuristic-based algorithms to provide their analysis. In other words, these kinds of models could predict player performance based on historical data, but accounting for more complex relationships between variables—e.g., injuries, team dynamics, and other external factors—wasn’t so much a thing.
In more recent years, AI has begun to emerge as a tool to enhance these predictive models and provide increasingly granular predictions and player-specific recommendations. With embedded features in fantasy apps, you can consult AI to automatically optimize your lineup based on predicted performance metrics, injury reports, and other advanced simulations that mirror real-world game scenarios (in an ideal world, of course).
Nowadays, while ML algorithms are processing massive amounts of historical data, AI-driven models—powered by real-time data from tracking systems (e.g., wearable devices and camera systems like the NFL’s Next Gen Stats)—are enabling more advanced predictions for fantasy owners looking to have a winning lineup. Take ESPN Fantasy, for example, which just recently teamed up with IBM to incorporate generative AI for enhanced trade analysis, details on player performance, and personalized recommendations.
Basically, the increasing use of AI to enhance predictive models has made fantasy sports even more data-driven and competitive, ultimately changing how many fantasy sports enthusiasts approach the game.
AI for fantasy football: Yea or nay?
For many, data analytics has always been a massive part of the allure of fantasy sports. It can be fun to compile and study a treasure trove of information on the team you’ve painstakingly drafted and put together for your season and then see the fruits of your labor pay off in the form of a “W” against your opponent.
So, does AI take away any of this fun? Maybe.
I conducted an informal survey of some longtime fantasy football enthusiasts to see where they were at (note—very informal, so don’t use this study for science 🧑🔬 ). When asked if they had ever used AI to to help them gather information or make decisions related to fantasy football, here were some answers:
“No. Should I?”
“Of course. But years ago.”
“Just the embedded IBM Watson analytics in the app, which is really more just like descriptive stats.”
And here are the results of a legitimate survey: IBM and Morning Consult recently polled 500 fantasy football users in the U.S. on whether they’re “embracing AI-powered solutions to help enhance their lineups and overall league performances.”
The results showed that nearly 90% (!) of users regularly engage with at least one AI-related tool; 82% said they make use of trade and waiver suggestions to improve their lineups; and of those not yet using AI, 92% said they would find it helpful.
This leads to the next idea: Can you put all of the decision-making in the hands of AI? And what happens if you do? 👀
Put me in, Coach AI
Back in 2011, IBM Watson defeated the two highest-ranked human Jeopardy! Players (at the time), widely showcasing the advanced capabilities of ML and natural language processing to the public.
Okay, so Watson can beat Ken Jennings, but how does he do as a fantasy football coach?
Spoiler: He can go undefeated through the regular season, apparently.
You can check out all of the results in a blog from IBM, but here’s the gist: A data scientist at IBM asked Watson to “learn the language of the NFL and each of its more than 400 players, and fantasy football.” It used AI to make “critical recommendations each week, predicting the best players to start games and the players to pick up off the waiver wire.”
“In week 13, for instance, Watson compared two different quarterbacks and found sentiment from news stories and social media postings favored one over the other,” the blog writes. “That Sunday, both quarterbacks got the win, but the quarterback with more positive sentiment blitzed his opponent—throwing five touchdown passes for 419 yards en route to victory.”
Although he ended up with a perfect 13-0 season, training Watson to conduct this type of analysis was “extremely complex,” according to the blog. The project still required humans to annotate “thousands of sentences and phrases” for Watson to learn, followed by several other projects and experiments IBMers conducted to refine models and build simulations—specifically, more than 1,000 of them for each of the top 400 players.
It’s unclear when this study was conducted exactly, but we know it’s got to be at least a few years old, considering it opens with a line about the Patriots having come off of a good season (😉).
Regardless, it shows that AI has been capable of being a pretty solid fantasy coach for some time now. But how easy is it now, with large language model (LLM)-based bots like ChatGPT and Bard (now Gemini) out in the wild?
Using LLMs for fantasy football decision-making
Let’s ask ChatGPT itself how helpful it is for making fantasy football decisions, straight from the horse’s mouth.
Here are the pros:
- Data insights: It can provide up-to-date player stats, team performances, and injury news to help inform your decisions.
- Matchup analysis: It can analyze player matchups, team defenses, and trends to recommend which players might perform better in a given week.
- Injury and news updates: It can help track breaking news about injuries or changes in team depth charts that could impact player performance.
- Trade evaluations: It can help evaluate potential fantasy football trades by assessing the value of players based on historical data and recent form.
- Lineup recommendations: Based on your league’s scoring system, it can suggest optimal starting lineups or waiver wire pickups.
And here are the cons:
- Real-time data: ChatGPT doesn’t have real-time access to live game stats, so you’ll still need to check player statuses—like injuries or last-minute roster changes—closer to game time (i.e., better check that app at 12:59 ET!).
- Human intuition: Fantasy football can have unpredictable elements, of course. While ChatGPT can offer data-driven insights, it may miss certain nuanced factors that human intuition or a gut feeling captures.
- Personalized knowledge: ChatGPT won’t know your league’s specific scoring system or the tendencies of your fellow managers (are they tough on approving trades?) unless you provide those details, so you’ll need to factor that in yourself.
In other words, ChatGPT sounds like a solid resource to get quick, informed advice or analysis. But for a winning strategy (and all the marbles and bragging rights): You’re still going to want real-time updates, opinions from the experts, your own league-specific context, and confidence in your own decision-making skills.
For example, I just asked ChatGPT, “Who should I start in fantasy football this week, Travis Kelce or Aaron Rodgers?” (If you’re not a football person: Kelce is a tight end, and Rodgers is a quarterback, so this question is a bit illogical. Unless you have a very bizarre setup in your league—hey, you never know—you’re not going to be pitting these two against each other in terms of whom to start in your lineup.)
I’ll give props to ChatGPT for letting me know right away that these two do, indeed, play two different positions so you don’t really need to directly compare them. But it still goes on to suggest I check the matchups for both players, monitor their injury reports, and look at the overall offensive performances of both the Chiefs and the Jets. In the end, it suggests that Kelce is the more reliable fantasy option, but if Rodgers has a favorable matchup and is in good form, it could swing in his favor.
But what about for drafting decisions?
There’s an interesting Reddit post entitled “I Let AI Call the Shots in My Fantasy Football Draft, Here’s How It Went,” which details (at a very intricate level) how Google Bard—now Gemini—was used to draft players for a 2023 fantasy team.
This person trained Bard by first providing basic pieces of fantasy football advice (e.g., don’t draft a quarterback in the earlier rounds) and teaching it to favor unique players, but it still failed to do things like take high-potential rookies into account (because there isn’t historical data on them) or consider the downside of aging players (because it is looking at historical data, not the impending joint deterioration of a 30-something-year-old).
But after giving it more time and training (and understanding its limitations), this person said Bard became “much easier to work with” and ultimately called its drafting choices “not great, but not bad either.”
TL;DR: Don’t take chatbots’ answers too seriously as a coach, especially if you’re a newbie with limited football knowledge. Using ChatGPT or Gemini as a supplementary tool could be beneficial, however, if you’re stuck in a bind and need suggestions that’ll look at the straight-up facts (as long as you provide relevant prompts).
The future of AI in sports analytics
Whether you’re a fan or not, AI-powered insights for sports are here to stay, whether it’s through widely accessible bots or embedded capabilities within every mainstream fantasy app.
But despite what AI is capable of (i.e., making all of your fantasy decisions for you), many fantasy fans are likely still leaving the final decisions up to their own expertise, simply because it’s part of the whole experience as an owner. It’s fun!
After all, human umpires are still a part of the game of baseball, despite the fact that every pitch is automatically being analyzed as a ball or strike, regardless of the home plate umpire’s ultimate decision (which cannot be challenged). Human decision-making (and thus, error) is simply still a part of sports.
Also, let’s not forget how truly unpredictable (to an oftentimes very frustrating degree) sports can be. ChatGPT isn’t going to help you predict a season-ending achilles tendon rupture to your quarterback within the first few plays of a game in Week 1. Gemini isn’t going to know your entire team—although expertly picked according to historical data—is going to be a total bust this year. It happens.
Likewise, we at Tellius have always advocated for a human-in-the-loop approach to integrating AI analytics in general. While AI can, of course, automate critical tasks in data analysis, it doesn’t replace the need for human involvement. Analytics experts still need to define metrics, formulate the right questions, interpret the results, and make strategic decisions—whether that’s for your business or your imaginary football team.
You’ll always have a place, coach! 💁