What 9,282 World Cup Predictions Tell Us About Football Fans

During the World Cup 2026 beta, 118 Pickz players predicted the scoreline of every one of the tournament's 104 matches — 9,282 player-made predictions in total, from the opening group game to Spain lifting the trophy. That's a genuinely interesting dataset, because it captures something rarely measured: how ordinary football fans actually predict matches when nothing is riding on it but pride.

So we analysed all of it. How good are football fans at calling matches — and where do they consistently go wrong? The short version: fans are better at picking winners than you might expect, far worse at exact scores than they believe, and there is one type of result they systematically refuse to predict. This article is the first in our prediction data and research series, built entirely on first-party Pickz data (full methodology below). It pairs naturally with our companion piece on what the tournament taught us about predicting, which covers the lessons rather than the numbers.

World Cup predictions in numbers

• 9,282 player-made predictions across 104 fixtures
• 118 predictors over the full tournament
• 57.0% of predictions called the correct result
• 900 predictions — 9.7% — nailed the exact score
• 26.3% of all predictions were 2-1, one way or the other
• The crowd's majority verdict was right in 72 of 104 matches (roughly 69%)
• 21 private leagues ran across the tournament

One important note before we start: these figures cover player-made predictions only. Where a player missed a deadline and a scoreline was auto-filled for them, that record is excluded from every behavioural statistic in this article — auto-fill tells you about the product, not about how fans predict. And this wasn't a crowd of casual dabblers: the thresholds below nest inside one another, and they show how deep participation ran.

How accurate was the Pickz crowd?

Accuracy sounds like one number. It's actually three, and keeping them separate is the key to reading everything that follows.

Crowd consensus accuracy. Take each fixture, find the outcome most players backed, and ask whether the majority was right. By this measure the Pickz crowd correctly called 72 of 104 matches — roughly 69%. The wisdom-of-crowds effect is real: pooling everyone's verdict beats almost any individual.

Individual correct-result accuracy. Across all 9,282 player-made predictions, 57.0% got the match result right — home win, draw or away win. That's the honest measure of a single fan's judgement, and it's meaningfully lower than the crowd's 69%. To be clear: it is not the case that 69% of individual predictions were correct. The crowd is smarter than the people in it.

Exact-score accuracy. Just 900 predictions — 9.7% — landed the precise scoreline. A fan who calls the right result more often than not will still only nail the exact score about once every ten matches. A result covers many possible scorelines; an exact score is one line among them, and the gap between 57.0% and 9.7% is the whole difficulty of score prediction in two numbers.

Fixture by fixture, that consensus record splits cleanly: the majority verdict was right on 72 fixtures and wrong on 32.

Football fans have a draw problem

Here is the finding that runs through this entire dataset. Pickz players predicted a draw in 18.1% of their predictions. Actual matches finished level 23.1% of the time. That's a five-percentage-point gap — fans under-selected the draw relative to how often it actually happened.

Why do fans avoid the draw? The sample can't prove motive, but it invites some plausible hypotheses. Picking a winner may simply feel more like predicting; a draw can feel like sitting on the fence. A 1-1 may look less satisfying than a 2-1 that says something about the match. And perceived favourites may soak up predictions that a colder read would have marked level. Whatever the cause, the pattern among Pickz players was consistent — and as we'll see, it's also where the crowd's most expensive mistakes lived.

Why does everyone predict 2-1?

Of 9,282 predictions, 26.3% were 2-1 — one scoreline, in one direction or the other, accounting for more than a quarter of everything submitted. It's easy to see the appeal. A 2-1 says: I've picked a winner, I think both teams will score, and I expect a real contest. It feels precise without feeling reckless, and it's a common enough result that it never looks naive. But when a quarter of all predictions crowd onto a single scoreline, plenty of them are habit rather than judgement. If that habit sounds familiar, our guide to how to predict football scores is about replacing the reflex 2-1 with an actual read of the match.

The match almost everyone got right

Crowd consensus at its best: Paraguay 0-1 France. An almost unanimous 98% of players backed the French to win, and they did — narrowly, but never in much doubt. When a fixture offers a clear favourite in decent form, a hundred-odd independent judgements converge on the truth with striking confidence.

The prediction that fooled almost everyone

Then there's Germany against Paraguay in the round of 32. Ninety-nine per cent of players backed Germany. Paraguay won on penalties — and because Pickz scores a shootout win as a win for the side that lifts the tie, not a single point was awarded to anyone on that fixture. Not one. The same crowd that read Paraguay 0-1 France almost perfectly was, one round earlier, as close to unanimous as it ever got — and completely wrong. Strong consensus measures confidence, not correctness.

Why draws caused the biggest prediction problems

Look at the five fixtures where the largest share of the crowd got the outcome wrong, and one thing jumps out: all five were draws. Every single one of the crowd's biggest collective misses was a match that finished level. That closes the loop on the draw problem — fans didn't just under-select draws in general; the draws they failed to see coming were precisely the results that cost the most points. In this sample the under-predicted outcome and the most punishing outcome were the same thing.

Were knockout matches easier to predict?

In this Pickz World Cup sample, yes — and by a wide margin. Player predictions called the correct result in 64.8% of knockout fixtures against 54.6% in the group stage.

A ten-point gap deserves caution before it becomes a theory. Knockout fixtures were fewer, the strongest sides had largely survived to contest them, and by the later rounds players had a month of form to work with instead of pre-tournament guesswork. Group stages also serve up more mismatched fixtures with rotated line-ups and dead rubbers. So we wouldn't claim knockout football is universally easier to predict — but in this tournament, for these players, it clearly was.

Did the crowd call the champion?

When it mattered most, yes. In the final, 75.9% of predictions — 41 of 54 — backed Spain to beat Argentina, and Spain won it 1-0. One round earlier the crowd had been burned in exactly the opposite way: 76% backed England in their semi-final, and Argentina won 2-1. The two results side by side are the whole story of crowd prediction — emphatic, usually right, and never safe.

One prediction nearly half the crowd nailed exactly

Exact scores run at 9.7% across the tournament, which makes this outlier remarkable: for the quarter-final between Norway and England, 48.4% of players predicted 1-2 — the exact final score. Nearly one in two players called not just the winner but the precise scoreline, on a knockout fixture, five times the tournament-wide exact-score rate. When a match sets up cleanly — a favourite expected to win a tight, competitive game away from home — fan intuition can be collectively sharp to a degree the overall averages hide.

What did we learn about football predictions?

• Fans are decent at picking winners. A 57.0% individual correct-result rate across three possible outcomes is genuinely respectable, and the pooled crowd verdict reached roughly 69%.
• Exact scores are another sport entirely. At 9.7%, even good judges of a match nail the scoreline about once in ten attempts.
• Draws appear systematically under-selected — 18.1% predicted against 23.1% actual in this sample — and every one of the crowd's five biggest misses was a draw.
• 2-1 exerts a pull on predictions out of proportion to any read of the fixtures, at 26.3% of everything submitted.
• Consensus is not correctness. The same crowd hit 98% on Paraguay 0-1 France and scored zero points on Germany–Paraguay after backing Germany 99 to 1.

One tournament, one (unusually engaged) group of players — the numbers here describe this sample, not football fans everywhere. But the patterns are coherent, they echo what prediction research has long suspected about draw-aversion, and they give us a baseline. As Pickz runs across the Premier League season, we'll be able to test which of these behaviours hold up over 38 gameweeks — if you play in a private league, our prediction league tips guide covers how to use exactly these tendencies against your mates.

Methodology

This analysis uses first-party Pickz data from the World Cup 2026 beta (11 June – 19 July 2026). The dataset covers 118 predictors and 9,282 player-made fixture predictions across all 104 tournament fixtures, plus tournament picks (including champion selections) made separately before the tournament. 2,878 auto-filled prediction records — scorelines entered automatically when a player missed a deadline — are excluded from every behavioural statistic.

• Correct result: the predicted outcome (home win, draw or away win) matches the match outcome. Matches decided by a penalty shootout count as a win for the side that progressed.
• Exact score: the predicted scoreline matches the final score. For shootout matches, the score at the end of play is what's compared.
• Crowd consensus: the outcome backed by the largest share of players for a fixture, judged right or wrong against the actual outcome.
• Percentages are rounded to one decimal place; the 69% consensus figure is 72 of 104 fixtures.
• Prediction volume varies by fixture, so per-fixture percentages rest on different sample sizes.
• Limitations: this was a warm, invited beta audience of 118 predictors, not a random sample of football fans, and a single tournament. Treat the patterns as observed behaviour in this sample rather than universal laws.

More guides