AI has tipped James Acaster to win The Celebrity Traitors, according to new analysis from leading search and data innovation agency Varn Search Marketing.
Varn analysed the contestants in two stages. First, AI was used to conduct an entity analysis of every celebrity, building a profile of their publicly documented traits, personality and characteristics.
Those profiles were then put to three leading AI models (ChatGPT, Gemini and Claude), which were each asked to rank the contestants from most to least likely to win The Celebrity Traitors. The AI was asked to base their decisions solely on those characteristics, rather than spoilers, betting markets or knowledge of what happened during filming.
Despite working from broadly the same information, the three models reached strikingly different conclusions about who was best equipped to survive the castle. Varn then combined the three sets of rankings to create an overall AI prediction, calculating each contestant’s average position across the models.
And the machine consensus puts comedian James Acaster in first place, narrowly ahead of singer and former Army officer James Blunt. Actress Sharon Rooney places third, followed by Jerry Hall and Maya Jama and Rob Beckett.
At the other end of the table, the combined AI rankings place Richard E. Grant, Myha’la and Ross Kemp last.
The entity analysis characterised Acaster as imaginative, quirky and self-deprecating, but also highlighted something potentially more useful in The Traitors: “meticulous preparation behind a casual exterior”, alongside sharp observational skills and a deadpan style.
Those characteristics helped him achieve the strongest overall position once the rankings from all three AI models were averaged.
While Acaster has the backing of the combined AI wisdom, ChatGPT predicted Joe Lycett will emerge victorious, with Gemini backing Hannah Fry and Claude picking Miranda Hart.
Andy Mollison, head of search and innovation at Varn, said: “The Traitors is a fascinating test for AI because winning isn’t simply about being the cleverest or most strategic person in the room. Contestants have to read people, build trust, manage how others perceive them and know when to speak up – or keep quiet.
“What interested us most wasn’t actually that AI ultimately picked James Acaster. It was how differently the individual models interpreted exactly the same group of people.
“Each AI model selected a different contestant as its most likely winner, which is a useful reminder that there isn’t really one single ‘AI opinion’. Different models can take the same underlying information and reach very different conclusions.
“By averaging the three rankings, we’ve created something closer to a collective AI prediction. Now we get to see whether the machines actually understand the characteristics needed to survive the castle – or whether humans remain considerably harder to predict.”
