By Paul Morris, founder of Guess a Game
This summer I ran a free score predictor for the World Cup at guessagame.com. 131 players called all 104 games — 11,658 predictions in total, every one of them settled against the real results. Now the confetti has been swept up in New Jersey, the data tells a story about how football fans actually predict: what we get right, where we kid ourselves, and what “good” really looks like in a prediction league.
If you play score predictors — Super 6, Kicktipp, a spreadsheet with your mates — some of these numbers might change how you pick.
The one-in-eleven rule
Across the whole tournament, 9.1% of predictions nailed the exact score. That’s the honest baseline: play a full World Cup and roughly one pick in eleven lands perfectly. Calling the result (win, draw or defeat, 55.3%) is only a little better than a mildly educated coin flip.
The spread between players is where it gets interesting. Among the 107 players who predicted at least 80 games, the best result-caller hit 70% and the worst managed 28% — worse than picking home wins every time. Our exact-score champion landed 20 perfect scores from 103 games, nearly one in five. Over a hundred-game sample that’s not luck; that’s someone who understands what scorelines football actually produces.
Which brings us to what most of us get wrong.
The 2-1 reflex and draw blindness
The most-predicted scoreline in our data was 2-1 — picked 1,491 times, ahead of 1-1, 2-0 and 1-2. Fans love 2-1: it feels specific, it hedges towards excitement, it says “I think they’ll win but I’m no mug.”
Meanwhile, 28% of group games ended in a draw — but only 21% of predictions were draws. Almost every player in our league systematically under-picked the most boring outcome in football, in a group stage that averaged 2.99 goals a game. If you take one number from this article into your own prediction league, take that one: you are probably not picking enough draws either.
The games nobody saw coming
The beauty of settling every prediction is you can rank the tournament’s shocks by how few people called them:
• Spain 0-0 Cabo Verde — called by 1% of players. The eventual champions, held goalless by the tournament’s smallest nation. The single most unpredicted result of the World Cup.
• Uruguay 2-2 Cabo Verde — 3%. Cabo Verde again. No one learned.
• Qatar 1-1 Switzerland — 5%.
• Portugal 1-1 DR Congo — 5%.
• England 0-0 Ghana — 6%. England fans will not need reminding.
• Ecuador 2-1 Germany — 6%. The result that detonated a thousand brackets.
Notice the pattern: five of the six biggest shocks were draws. Draw blindness isn’t just a leak; it’s where prediction leagues are won and lost.
At the other end, the most predictable games of the tournament: 97% of players called France beating Iraq (3-0), 96% had Spain over Saudi Arabia (4-0), and 95% backed Germany against Curaçao — though precisely nobody suggested it would finish 7-1.
Knockout football is (slightly) easier to predict
You might expect knockout games between well-matched sides to be harder to call. The opposite happened: exact-score accuracy rose from 8.4% in the groups to 10.8% in the knockouts. Once the minnows go home, football gets more conservative and scorelines cluster — 1-0s and 2-1s — which is exactly where predictors live. (The France–England third-place game, a ten-goal 4-6 lunacy won by England, is the exception that proves it.)
The final was the coin flip it deserved to be: 47% of players had Spain beating Argentina, and just 7% called the exact 1-0. The most popular picks were 2-1 and 1-2 — that reflex again.
Then the new season started, and everyone got humbled
Guess a Game now runs season-long predictors for the Premier League, English Championship and Scottish Premiership, so the World Cup class of 2026 has rolled straight into the 2026/27 season. Early verdict: club football in August is chaos.
In the Championship’s opening round — eleven games at 2.91 goals each — our players called just 24% of results correctly, with two exact scores between them. That’s half their World Cup accuracy. Portsmouth 1-3 QPR, Wolves 2-2 Blackburn, four different 2-1s: promoted, relegated and rebuilt squads that nobody has properly seen yet.
The Scottish Premiership has been wilder still: St Johnstone 4-3 Kilmarnock, Kilmarnock 1-5 Celtic a week later, Hearts 4-0 Dundee United. Six home wins, two draws and four away wins from twelve games.
We also run a stats model — The Bot — that publishes its prediction for every game before kick-off. It called 36% of Championship results and 33% in Scotland, beating the humans, but it hasn’t landed a single exact score yet. August football makes fools of everyone, silicon included.
There’s a lesson in that for anyone starting a season-long prediction league: the first month is a lottery, so a league that rewards consistency over 46 gameweeks will always find a better champion than one decided by a hot fortnight. Don’t let anyone quit in September.
What good looks like
Pulling it all together, the data suggests a realistic scorecard for a serious prediction-league player over a big sample:
• Exact scores: 10% is solid, 15% is excellent, 20% is elite.
• Results: 55% is par, 65%+ wins most leagues.
• Draws: if fewer than a quarter of your picks are draws, you’re leaking points.
• Scoreline choice: 2-1 feels clever; it usually isn’t. The modal football score is still 1-1 or 1-0.
And the meta-lesson from 11,658 predictions: nobody — not the 70% result-caller, not the bot, not the bloke who got twenty exact scores — saw Spain 0-0 Cabo Verde coming. That’s why we play.
Guess a Game is a free, non-betting score predictor for the Premier League, English Championship and Scottish Premiership — predict every game, build your final table, and run a private mini-league with your mates at guessagame.com.