Dr. Angel Torres-Toukoumidis
Full Professor at Universidad Politécnica Salesiana, Ecuador, and Editor-in-Chief of Universitas-XXI. He is Coordinator of the Gamelab-UPS Research Group, holds a PhD in Communication and a Master’s in Communication for Social Purposes, and has authored more than 200 scholarly publications on communication, educommunication, gamification and digital culture.
Email: atorrest@ups.edu.ec
Every World Cup has demanded an oracle capable of allaying uncertainty: Paul the Octopus discharged that function from his tank in South Africa 2010, followed by Cabeção the turtle in 2014, Achilles the cat as his successor in 2018, and ultimately Taiyo the otter in 2022. By 2026 that figure has been supplanted by an array of Artificial Intelligence (AI) models: Gemini, Claude, GPT, DeepSeek, Qwen, Grok, among others, which journalists and supporters alike interrogate with one and the same question: who will lift the World Cup on July 19th in the United States? The communicational logic remains identical to that of Paul, Achilles, or Taiyo, inasmuch as an external figure is sought to transmute sporting uncertainty into narrative; the novelty lies in the fact that the new oracle bears a technical guise. A closer examination of how these responses are generated suggests that AI-based World Cup predictions are not neutral computational outputs but journalistic narratives shaped through prompting and editorial selection.
La Nación drew on Gemini, grounding it in Opta statistics and market odds, and settled upon Spain, even though Infobae recorded that, in a separate trial, the very same model wavered between Spain, Argentina, and England. Claude, consulted by Olé USA and Página 12, concurred in Spain defeating France. GPT, by contrast, was consulted by AVPasión and La Nación, yielding Brazil as champion, whereas El Cronista directed one and the same prompt at Gemini, ChatGPT, and Claude, which coincided in Spain. DeepSeek, for its part, insistently reiterates Argentina, as evidenced in Tiempo and Decrypt (18%). Such predictions arrive distilled into a single figure within a table.
The crux of this divergence resides in the prompt and in the editorial decision as to which verdict warrants publication. There exists no single answer from the AI, but rather results conditioned by the architecture, the available data, and the wording of the question. The very name of the model operates, moreover, as a seal of authority, seasoning the argumentum ad verecundiam: to assert that “ChatGPT, or Claude, or DeepSeek predicted it” suffices to invest with objectivity a text whose precise formulation is seldom rendered transparent. The consultation of AI thus functions as a reproducible formula that any newsroom may execute in order to manufacture, at minimal cost, a piece of news bearing the semblance of revelation.
This formula has become serial. Decrypt consulted seven agents in order to assemble a ranking, while ScoreGPT pledges to update daily the consensus of six models, thereby converting the forecast continuously updated content; rather than rendering explicit the disagreement among systems, the genuinely informative datum, the coverage resolves it into a single headline-ready figure. More rigorous endeavors do exist: the “AI Prediction World Cup” run by Financial Times Alphaville pits Gemini, Claude, several versions of ChatGPT, a Goldman Sachs model, and human projections against one another, thereby confirming that these systems are no single, infallible oracle, but rather disparate competitors whose performance varies match by match.
It thus follows that outlets such as El País, Forbes Centroamérica, Debate, Covers, and The Mirror have transformed the AI consultation into a serial formula, low in cost and high in clicks. The phenomenon becomes particularly pronounced when The Sun requested that Gemini act as a “Nostradamus” of the World Cup and dramatize its forecasts England crowned champion, a final against Argentina – the very framing betrays that AI is deployed as a narrative device rather than as sound sporting evidence.
On social media this logic grows radicalized, for it is there that the value of the forecast resides in its virality. Such content thrives by conjoining three ingredients of the highest viral potential: football, the uncertainty of the outcome, and the fascination exerted by AI. A confluence that invites the sharing of prompt screenshots, rankings of favorites, and contradictory predictions among rival fan bases. The authority of the verdict proceeds neither from a transparent methodology nor from a declared margin of error, but rather from the symbolic prestige ascribed to the model: the logo of a recognizable AI alone suffices for the screenshot to be forwarded as though it were a datum and not a conditioned statistical output. In this manner, the platform rewards instantaneous tribal identification over verification, and each act of forwarding reinforces the illusion that a closed answer exists where there is, in truth, but a conversation with a machine.
In closing, it is fitting to dwell upon the communicational dimension of the phenomenon: what the public receives is not the prediction of an AI, but rather its journalistic edition. Between the model and the reader there intercede invisible decisions, which prompt was used, which response was selected, which headline was drafted, that transmute a probabilistic output into a spectacular verdict. Sports media continue to seek authoritative voices capable of reducing uncertainty, and AI provides an endlessly available oracle wrapped in technological credibility.upo
