Por Renan Filho · 2026-09-27The NFL betting community has watched with a mix of amusement and skepticism as ChatGPT, the flagship large language model from OpenAI, attempts to outpick human experts. Through the first two weeks of the 2026 season, the results have been brutal: the AI sits at a dismal 2-8 against the spread (ATS), while the human panel—a group of veteran sports bettors and statisticians—has posted a respectable 6-4 record. But Week 3 offers a fresh slate, and the question is no longer whether ChatGPT can be perfect, but whether it can avoid total irrelevance.
This experiment, run by a prominent sports analytics site, pits a carefully crafted prompt against a rotating team of human analysts who publish their picks every Wednesday. The rules are simple: each side picks 10 games against the spread, and the results are tracked transparently. The early data has been damning for the AI. In Week 1, ChatGPT went 1-4, missing key spreads like the Kansas City Chiefs laying 7.5 points to the Houston Texans (who covered easily) and the Dallas Cowboys losing outright as 3-point favorites. Week 2 wasn’t much better—ChatGPT’s only bright spot was correctly predicting the San Francisco 49ers would cover against the Seattle Seahawks. Meanwhile, humans correctly faded the public on the New York Jets, who upset the Tennessee Titans as 4-point underdogs.
Why has the AI struggled? According to a preliminary analysis from the tracking site, ChatGPT appears to rely too heavily on recent narratives and broad statistical averages, ignoring the granular, often irrational factors that sway NFL games: injuries not reflected in box scores, weather forecasts that shift lines, and the “money” moves from sharp bettors that move markets after the initial lines are set. Humans, by contrast, use a mix of quantitative models and qualitative knowledge—like knowing that a kicker is struggling with the new wind patterns at a particular stadium, or that a linebacker’s personal turmoil might affect his focus. This aligns with a core liberal-conservador principle: decentralized, emergent knowledge (what Hayek called “the use of knowledge in society”) often outperforms centralized, top-down computation.
But let’s not write off the AI entirely. The human panel’s 6-4 start is solid, but not spectacular. Over a full season, even the best handicappers hit around 52-55% ATS. Luck plays a significant role in small samples. And ChatGPT has one advantage: it does not panic. While human bettors can become emotional after a bad week, doubling down or chasing losses, the AI will apply the same logic regardless of past performance. If its underlying data and reasoning are sound, regression to the mean could bring it back to respectability. In Week 3, the AI’s picks include taking the Baltimore Ravens as 6-point favorites at home against the Indianapolis Colts—a popular pick among models due to Lamar Jackson’s dominance over AFC South teams—and betting against the New England Patriots, who are 3-point road underdogs to the New York Giants. The humans, meanwhile, are leaning on the Green Bay Packers to cover against the Chicago Bears, citing the Bears’ offensive line injuries.
From a free-market perspective, this competition is a beautiful demonstration of how decentralized human judgment can still compete with, and even beat, centralized algorithmic predictions in a complex, adaptive environment. The NFL is not a controlled laboratory; it is a chaotic system of 22 moving bodies, each with human emotions, contracts, and incentives. Artificial intelligence may excel at chess or Go, where the rules are fixed and the state space is finite, but sports betting remains a domain where local knowledge and real-time intuition often carry the day. That doesn’t mean AI is useless—it can process mountains of data far faster than any human—but it must be paired with human context to be truly effective. The best sports betting models already blend machine learning with human adjustment; ChatGPT, as a generalist, lacks that fine-tuning.
The liberal-conservador editorial line here should not be interpreted as anti-technology. We embrace innovation and free markets; the AI development is a testament to human ingenuity. But we must also recognize the limits of any top-down system. If ChatGPT continues to post losing records, it will be yet another reminder that machines are tools, not oracles. The market—in this case, the betting market—is far more intelligent than any single algorithm because it aggregates the dispersed knowledge of thousands of participants. The lines themselves move based on the collective wisdom of sharp and public bettors alike. Beating that aggregate is extraordinarily difficult, and ChatGPT has thus far failed.