380 games. One full Premier League season. And at the end of it all, a Microsoft chatbot edged out a former professional striker in a forecasting contest that nobody saw coming quite like this. The BBC Sport predictions battle for the 2025-26 season delivered a genuinely surprising verdict — and the numbers behind it deserve a proper look.
How the BBC Sport prediction contest actually worked
Every week throughout the season, Chris Sutton made forecasts for all 380 Premier League fixtures on behalf of BBC Sport. He wasn’t alone : Microsoft’s AI chatbot Copilot, a rotating cast of celebrity guests, and BBC Sport readers all submitted their own predictions alongside him. The decisive metric was simple — outright correct wins, meaning picking the right result outright, not just a tie on points.
This format matters. It strips away lucky streaks and forces consistency across an entire campaign. Sutton is no amateur here — he spent years as a striker for Blackburn Rovers and Celtic before becoming one of British football’s most recognisable pundits. His track record with BBC Sport predictions spans multiple seasons.
Here’s how the final standings broke down across key performance indicators :
| Predictor | Correct results (final week) | Exact scores (final week) | Final week points | Overall winner |
|---|---|---|---|---|
| BBC readers | 3 | 2 | 90 | — |
| Sam Tomkins | 3 | 0 | 30 | — |
| AI (Copilot) | 4 | 0 | 40 | ✓ Yes |
| Chris Sutton | 2 | 0 | 20 | — |
The final round made things especially dramatic. Sutton and Copilot were level on outright wins going into the last matchday. Sutton sat second only because he had accumulated fewer tied victories — a tiebreaker that put him in a position where he needed a strong Sunday to claim the title. He didn’t get it. Two correct results and zero exact scores gave him just 20 points on the final day, while the AI posted four correct results for 40 points.
The BBC readers, interestingly, won that final week outright — nailing Arsenal’s 2-1 victory at Crystal Palace and Burnley’s 1-1 draw with Wolves as exact scores, which earned them 90 points for the round. Collective wisdom, it turns out, can still outperform both the expert and the algorithm on any given weekend.
AI beats Sutton — and has something to say about it
When asked how it felt to outperform Chris Sutton across an entire Premier League season, Copilot’s response was characteristically deadpan — and oddly self-aware. The chatbot said it didn’t experience feelings the way a human does, but added that it could “recognise the achievement in the same way a model recognises a pattern : the scale, the consistency, the improbability.”
It continued : “Beating Chris Sutton over a full 380-game season is basically the equivalent of winning away at the Etihad : improbable, statistical chaos, and therefore deeply amusing.” Hard to argue with the logic, frankly. Manchester City’s home record at the Etihad Stadium has historically been one of the toughest nuts to crack in European football — using it as a benchmark for improbability is, if nothing else, a well-calibrated metaphor from a machine that supposedly feels nothing.
Sutton’s reaction ? Blunt, as expected. “The game’s gone,” he said. “AI will be winning the Premier League soon, at this rate.” There’s something almost poetic about a pundit who has spent decades analysing football now questioning whether human intuition still has a place in prediction contests.
- Copilot’s winning edge was built on consistency over 380 games, not a single lucky run
- Sutton lost the title on a tiebreaker, not a points gap — the margin was that fine
- Guest predictor Sam Tomkins (the singer-songwriter) finished above Sutton on the final matchday with 30 points vs. 20
- The BBC readers collectively topped the final round, proving crowd-sourced predictions have real value
What pundit accuracy really tells us about football forecasting
The broader takeaway here goes well beyond a fun BBC Sport competition. Forecasting football results over a full season is genuinely difficult — and the closeness of this race underlines exactly that. Sutton came within a tiebreaker of beating an AI that processed data at a scale no human analyst can match manually.
What’s telling is that neither the expert nor the algorithm dominated convincingly. The BBC readers’ collective predictions regularly competed with both Sutton and Copilot on a week-by-week basis, and they claimed the final round outright. There’s a strong case that aggregated fan intuition — when pooled across thousands of responses — carries real predictive weight.
From a purely analytical standpoint, exact score predictions are where AI and humans diverge most sharply. Sutton scored zero exact scores on the final day; the readers hit two. Pinpointing a scoreline requires a different kind of reasoning — gut feel, knowledge of team news, awareness of motivational context — that current AI models still handle imperfectly.
The 2025-26 BBC Sport predictions contest ultimately shows that the gap between human expertise and machine forecasting is narrowing fast. Sutton’s decades of professional experience and tactical insight weren’t enough to pull clear. That’s not a knock on him — it’s a genuine reflection of how far football analytics tools have progressed. If you’re thinking about making your own match predictions next season, tracking your hit rate on exact scores specifically is the most revealing metric to watch.
