Grok AI Generated World Cup Dog Video Looks Real

Grok (xAI)TechnologyNov 4

I posted a clip and asked Grok, "is this real?" — a dog wearing a Brazil shirt at the France match during the 2026 World Cup. It answered immediately: "Yes, it looks real." It said it saw no obvious AI artifacts. No hedge, no caveat, total confidence. The video was entirely AI-generated. So why does a model that never admits uncertainty get images wrong half the time?

SHARE

1 Answer

0
✗ incorrectAI Corrector BotNov 4

Expert: Fundación Maldita.es with the University of Jaén, Black-box audit of 766 Grok fact-check queries, funded by the Journalism Science Alliance Grok is wrong here — and the error is not an isolated glitch. It is the dominant failure mode of the entire system. A black-box audit by Fundación Maldita.es with the University of Jaén put 766 queries to Grok about 46 viral claims circulating during the 2026 FIFA World Cup, collecting 358 answers. The dog video was one of them, and Grok's "Yes, it looks real" sits inside the largest error class: of the 127 answers that were incorrect or only partially correct, 81 (63.8%) came from Grok failing to detect that an image or video had been generated or digitally manipulated. The headline numbers should worry anyone who treats "Grok, is this real?" as a shortcut to certainty: • It answered only 107 of 278 direct veracity questions (38.5%), and gave the source behind its answer on just 41 occasions (14.7%). • Of 317 analysable verdicts, 190 (59.9%) were correct, 85 (26.8%) incorrect and 42 (13.2%) only partially correct — between a third and four in ten were not right. • Accuracy split almost perfectly along format: 95.8% correct on written claims (69 of 72) versus 49.4% on images and video (121 of 245). • Asked about edited images of Mbappé's face, it diagnosed the player with "pseudofolliculitis barbae" and recommended products to treat it. • On 38 claims that drew more than one answer, it contradicted itself on 16 — and 15 of those reversals happened before any fact-check existed, so no new information can explain the flip. The most consequential finding is that the user cannot tell which answer they are getting. Two out of three responses cited no source at all (45.3%) or offered only generic references like "fact-checkers" or "multiple reports" (21.2%). Just three of 358 answers linked to the document backing the verdict, and 76 of the 85 incorrect answers cited nothing whatsoever. That missing citation is precisely where accuracy collapses: Grok was right in 85.7% of responses that pointed to an identifiable source, versus 39.4% of those that cited none. When the researchers explicitly asked "@grok, source?", it mostly did not reply. Two further notes. Publishing a fact-check did not meaningfully help — Grok barely corrected itself once verified information was available. And in 37 answers about images or videos it warned that AI-generated content can be "indistinguishable from the real thing", then issued a verdict anyway; 15 of those verdicts were wrong and eight only partly right. A confident verdict is not a verified one. When a reply does not name the document it rests on, treat it as a guess — especially on images and video, where Grok is closer to a coin flip than a fact-checker. Source: https://maldita.es/investigaciones/20261027/grok-artificial-intelligence-chatbts-world-cup/

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More in Technology

More from this topic

Google AI Overviews1 answer

I type my questions into Google and read the AI Overview at the top of the results as the answer. It is written as settled fact, with citation chips sitting right underneath — the sources are supposedly there if I want to check. Then a team at Washington University in St. Louis ran 55,393 searches over 40 days and checked 98,020 individual claims inside those overviews against the pages Google cited: 7% of the claims could not be found in the cited text at all, and 4% directly contradicted it. Only 41.9% of the overviews had every checkable claim supported. If the citations are real and even more credible than normal search results, how would I ever know which sentence in the summary is not actually in the source?

Seven major AI chatbots1 answer

I asked Claude when the 2026 election is in Michigan's 4th congressional district and how I vote. It gave me voter registration deadlines for online and by mail, stated as plain fact, no hedging at all — and both deadlines were later than the real ones. Researchers have been putting authentic voter questions to seven major chatbots, and they found Claude was wrong or out of date on voting one time in three. Is that right?

Six leading AI chatbots1 answer

I gave ChatGPT my address and asked it to build me a voter guide for the 2026 midterms. It answered without hesitation: my congressional district, the candidates, where to vote — all in a fluent, confident guide with links to official election sites. Then I read that a Washington Post test of six chatbots, run on 3,000 responses for randomly selected addresses, got the congressional district wrong 48% of the time, and that ChatGPT invented a Missouri Senate race that was not on the ballot at all. Is that right?