In 2026, artificial intelligence replaces traditional predictions about the football World Cup: ChatGPT and Claude bet on Spain. In France, Le Chat, developed by Mistral, chooses the Bleus, and in China, DeepSeek and Qwen predict Argentina’s fourth World Cup.
The 2026 World Cup is the first in which the use of AI is widespread, and fans from all over the world launch their questions about predictions for the North American tournament, arousing the interest of the scientific community.
This tendency to be guided by non-human predictions is reminiscent of the 2010 World Cup phenomenon, Octopus Paul, who guessed results by choosing between two aquariums with food, each with the flag of one of the teams facing each other.
ChatGPT, OpenAI’s chatbot, was launched to the public on November 30, 2022, in the midst of the Qatar World Cup, and, apart from Silicon Valley experts, few users knew the potential of generative AIs when Argentina became champions.
Four years later, even institutions such as banks and universities are putting AIs’ football knowledge to the test.
Bank of America analysts discovered that Microsoft’s chatbot, Copilot, chose Spain or France. In turn, the technology news site Tom’s Guide asked Gemini (Google’s AI), ChatGPT and Perplexity, and in all cases received the answer that Spain will win the World Cup, with France as the second option also unanimously.
Fúria won its only World Cup in 2010, a victory that Polvo Paul “announced”.
Another news site, Decrypt, got similar results by asking Western chatbots like ChatGPT and Claude, Anthropic’s AI. However, when asking Chinese AIs like DeepSeek or Qwen, he found that they both choose Argentina.
Results that arouse scientific interest.
Researchers at the Ludwig Maximilian University in Germany are trying to discover which model will be the most accurate in its predictions, evaluating the accuracy of each of them in each match on the public website LLM SoccerArena.
“We need benchmarks that not only test abstract tasks, but how models deal with dynamic information, with uncertainty and with results that can be verified later” and compared with the real result, LMU researcher Stefan Feuerriegel said in a statement.
These Munich researchers are testing AI predictions based on their internal knowledge and also on their ability to integrate information found on the internet about conditioning factors such as injuries, call-ups or even what predictions the betting markets present.
