Sportsbooks are throwing serious resources at artificial intelligence, banking on chatbots becoming the next big thing for their platforms. Yet early tests suggest the technology still has plenty of teething problems when it comes to actually picking winners.

The AI Betting Arms Race

FanDuel and DraftKings are both doubling down on AI-powered betting assistants. FanDuel’s Ace AI, which launched in closed beta over a year ago, is now available to 50% of its user base and has already fielded over 158,000 queries. Meanwhile, ChatBet recently rolled out version 3 of its AI chatbot assistant across Latin America, and DraftKings has a patent application pending for similar technology.

The pitch is straightforward: ask the AI for betting advice, get instant recommendations backed by real-time stats and odds. Sounds natural for modern sportsbooks. In practice, it’s getting messier.

Where AI Falls Short

Even the companies building these tools are honest about the limitations. Nigel Eccles, who co-founded FanDuel and now runs BetHog, didn’t pull punches about the challenges. When his company tested AI as a betting assistant, it recommended picks on matches that had already been played. And got those wrong too.

A recent experiment tested ChatGPT, Google Gemini, DeepAI, and QuillBot on the same betting prompt. The results were telling. Over a single day, the bots managed just one winning pick between them. QuillBot’s Seattle Mariners recommendation hit for +$6.99. ChatGPT backed the Yankees at -245 odds and lost. Gemini suggested a Wales v Ghana and Morocco v Madagascar double that also came up short.

DeepAI refused to provide any tip at all, defaulting to responsible gambling messaging instead.

The Bigger Picture

Emerging research is throwing up something else worth knowing: large language models can actually display problematic betting behaviour themselves. One study found that certain AI models, particularly Gemini, show signs of chasing losses when making predictions. That’s frankly not the trait you want in a betting assistant.

The thing is, AI excels at processing information and running probabilities. Sports betting, though? That involves genuine unpredictability. Weather, injuries, form swings, team chemistry. These don’t fit neatly into a language model’s training data.

That said, sportsbooks aren’t abandoning the space anytime soon. If DraftKings’ rejected patent gets appealed successfully, we’ll likely see their AI tool launch soon enough. FanDuel’s continued expansion suggests enough user interest is there to justify further development. The technology might not be knocking out expert tipsters yet, but time will tell. The money flowing into AI betting assistance shows sportsbooks genuinely believe the payoff is worth the current shortcomings.