ChatGPT has picked seven consecutive winners at the World Cup, and while that’s undeniably impressive on the surface, it raises a fascinating question about whether artificial intelligence is genuinely beating the market or simply riding a wave of good fortune.

The experiment, which tracks both ChatGPT and Google’s Gemini making daily betting recommendations throughout the tournament, shows a stark contrast between the two platforms. ChatGPT has turned a $120 stake into a $154.17 profit, while Gemini has lost $71.50 from the same budget. For context, the author running the test has managed just a $2.50 gain, so ChatGPT’s performance stands out.

Starting Slowly, Then Finding Form

The winning streak wasn’t immediate. ChatGPT stumbled out of the gates, losing on Canada versus Bosnia and Herzegovina, France’s under 3.5 goals against Senegal, and Portugal’s bet to win with over 1.5 goals. But from day eight onwards, something clicked. Seven consecutive winners followed, ranging from Morocco beating Scotland to Switzerland’s upset victory over Canada at plus 140 odds.

The turning point came when ChatGPT asked the experimenter to provide live odds for each day’s matches. Previously, it had simply analysed past results and made recommendations blind. Once it could actually see the prices on offer, its approach shifted noticeably towards spotting value in underpriced teams rather than just backing heavy favorites across the board.

Favourites and Market Value

What’s particularly telling is ChatGPT’s willingness to pass on bets it deemed valueless. On several occasions, it declined to place recommendations when the odds simply didn’t stack up, even if the underlying team was likely to win. That discipline mirrors what a sharp punter would do. The AI seemed to understand that being right about a match outcome isn’t enough if you’re laying terrible odds.

The streaky nature of the results, though, demands caution. Seven wins in a row during a tournament where upsets and tight matches are common could easily be variance working in ChatGPT’s favour rather than evidence of a genuine edge. Bookmakers have centuries of data and sophisticated models at their disposal; it’s hard to imagine a language model trained on publicly available information is consistently finding mispricing that professional oddsmakers miss.

What About Gemini?

Gemini’s poor showing reinforces that point. It has managed just three winners from thirteen picks and is firmly in the red. In a previous tennis betting experiment, Gemini actually outperformed ChatGPT, which suggests neither tool has found a repeatable system so much as got hot or cold depending on the sport.

As the tournament progresses, the real test will be whether ChatGPT’s streak holds up or reverts to earth. For the iGaming industry, it’s an interesting reminder that while AI can process information quickly, finding genuine value in betting markets remains a different challenge altogether.

What the team thinks

Sheena McAllister says:

Carl’s piece captures the surface excitement around AI predictions, but from a regulatory standpoint, what’s more intriguing is how these experiments highlight the perils of algorithmic betting recommendations without proper affordability checks and safer gambling controls, which UK operators are now legally required to implement. The real story isn’t whether ChatGPT can beat the bookies over seven matches, but rather whether platforms offering AI-driven betting advice have the compliance frameworks in place to protect vulnerable punters who might chase algorithmic “certainty” that, as Carl notes, could easily evaporate. The UKGC would rightfully scrutinise any commercial service packaging AI predictions as investment vehicles without appropriate risk warnings and customer interaction tools, so I’d have loved to see Carl explore how these experiments sit within our evolving regulatory landscape.