← Home · Research · Published July 16, 2026

How accurate are football fans? 194,812 World Cup predictions, analyzed

Football fans predicted the correct match outcome 60.0% of the time and the exact score 10.4% of the time across the first 102 matches of the 2026 World Cup — but when a match ended in a draw, only 15.7% of predictions saw it coming. This report analyzes every real, pre-kickoff score prediction made in BeTeam's private prediction leagues, compares fans against a built-in random baseline, and shows when the crowd is right — and how badly it can be wrong.

All figures are aggregated and anonymized; no individual player data is published. Every number on this page is generated by a documented, reproducible pipeline. Aggregate data is downloadable (CC BY 4.0).

Key findings

1. Fans got the winner right 60.0% of the time — 1.7× better than chance. Across 194,812 predictions on 102 matches (Jun 11–Jul 15, 2026), 60.0% named the correct outcome (home win, draw, or away win). Random score predictions on the same matches — generated by the app when a member misses kickoff — got 34.6%.
2. Exact scorelines are hard: 10.4% correct, about 1 in 10. A further 12.5% of predictions had the right goal difference without the exact score.
3. Draws are football's blind spot. 24 of the 102 matches (23.5%) ended level, but fans predicted a draw in just 13–16% of cases. On matches that actually drew, only 15.7% of predictions were right — versus 73.3% on home wins and 74.1% on away wins.
4. The crowd beats the individual. The majority pick was correct in 67 of 102 matches (65.7%), while the average individual prediction was right 60.0% of the time.
5. Crowd confidence is well calibrated. When 75%+ of fans agreed on a result, the crowd was right 74.6% of the time (47 of 63 matches). When agreement was under 45%, it was right just once in six matches (16.7%).
6. The crowd drew level with a web-searching AI. On the 19 knockout matches where BeTeam also showed a Google Gemini prediction (grounded in live web search, locked before kickoff), fans and the AI each called 13 of 19 outcomes correctly (68.4%) and split their six disagreements 3–3. Fans were better on exact scores (3 vs 1); the AI's wins were both semi-final upsets.
7. 2-1 is the default prediction of football fans. One in six predictions (16.9%) was exactly 2-1. The three most popular scorelines — 2-1, 2-0, 1-2 — covered 40.8% of all 194,812 predictions.
8. Even near-unanimity can fail. 99.4% of fans picked Spain to beat Cabo Verde; the match finished 0-0 — the most lopsided miss of the tournament so far. Both semi-finals also went against the crowd.
9. Predicting early costs nothing. Group-stage predictions submitted more than 7 days before kickoff were right 60.3% of the time; predictions in the final hour, 59.8%. Waiting for team news did not measurably help.
10. Prediction leagues stay alive to the end. In 91.1% of private leagues with 3+ players the lead changed hands at least once (median: 4 lead changes), and going into the final, the top-2 gap in 96.3% of leagues is still smaller than the points available from one exact final prediction.

Overall accuracy — and the random baseline

Three accuracy tiers, measured on all 194,812 deduplicated fan predictions, compared with 195,347 random auto-generated predictions on the same 102 matches (uniform random 0–3 scores — see definitions):

Accuracy by tournament stage

Knockout football compresses outcomes (favorites meet underdogs; one match, everything at stake), so stage-by-stage accuracy swings widely. The quarter-finals were the most predictable round so far (78.0% correct outcomes, and a remarkable 24.5% exact scores); the semi-finals were the least (31.3% — both went against the crowd).

Interpretation note: stages with 2–8 matches measure those specific matches, not knockout football in general. The round-of-16 exact rate (2.8%) collapsed because several matches produced unusual scorelines; the quarter-final rate (24.5%) was boosted by heavily-predicted results landing exactly.

The draw blind spot

Fans structurally under-predict draws. In the group stage — where every fixture was known long in advance and no match can go to extra time — 27.8% of matches ended level, but only 13.7% of predictions called a draw. Averaged over the whole tournament so far, when a match did end in a draw, just 15.7% of its predictions were right; a majority (58.5%) had backed the first-listed team to win outright.

Fans were also mildly optimistic about goals — but less than the cliché suggests: group-stage predictions averaged 2.75 total goals per match against an actual 2.99. The bigger distortion is where the goals go: fans convert "probably close" into "narrow win" rather than "draw". 0-0 — how 7 of the 72 group-stage matches actually ended — was picked in just 1.15% of predictions.

Crowd consensus: right 66% of the time, and well calibrated

Collapse each match to its crowd consensus — the outcome most fans picked — and the crowd called 67 of 102 matches (65.7%), beating the average individual (60.0%). More useful: the level of agreement predicted how trustworthy the pick was.

The failures are memorable precisely because they're rare. The most lopsided misses so far:

*Level after extra time; decided on penalties. Prediction scoring treats shootout matches as draws — see definitions. Both semi-finals also beat the crowd: 69.7% picked France (lost 0-2 to Spain), and the England–Argentina plurality (43.2% England) fell to Argentina's 2-1 win.

Crowd vs AI: dead level through the knockouts

Every game's stats screen in BeTeam also shows an AI score prediction — Google's Gemini, prompted to search the live web for recent form and team news, then locked before kickoff. That feature only launched mid-tournament (July 9), so a stored AI prediction exists for 19 knockout matches — the Round of 32 onward — and no group games. On those 19 shared matches, the wisdom of roughly 1,900 fans per game and a web-grounded large language model finished exactly level.

The crowd and the AI disagreed on the winner in just 6 of the 19 matches — and split them 3–3. The AI's edge came late and specific: it correctly called both semi-final upsets — Spain over France and Argentina over England — plus USA's collapse against Belgium, exactly the results fan loyalty to the bigger name got wrong. The crowd took the earlier disagreements, including England's 1-2 win at Norway, which fans pinned to the exact score while the AI expected a draw.

Sample: 19 knockout matches (Jul 2 – Jul 15, 2026); the AI made one prediction per fixture (Gemini 2.5 Flash with Google Search grounding), while each crowd figure aggregates 1,793–1,999 independent fan predictions. This is a knockout-round comparison, not a whole-tournament one — there is no group-stage AI coverage. Full per-match crowd-vs-AI data is in the downloads. The comparison will grow as the AI covers more matches in future tournaments.

The scorelines fans actually predict

Fan predictions concentrate heavily on a handful of "sensible" scorelines. 2-1 alone accounts for one in six predictions; the top three cover 40.8% of everything submitted. The market inefficiency is visible in the hit rates: the crowd's favorite 2-1 landed 12.5% of the time it was played, while 3-0 — picked far less — hit 18.8%.

Per match, fans produced between 17 and 30 distinct scorelines (median 22). The most unanimous single prediction of the tournament: 46.1% of fans said Brazil vs Norway (round of 16) would finish exactly 2-1. The most divided match was Iran vs New Zealand, where the most popular score (1-1) had just 20.2% support. Across all 102 matches, the single most-predicted scoreline was exactly right only 13 times (12.7%).

Does predicting early hurt?

No — at least not at this World Cup. Group-stage predictions made more than a week before kickoff performed as well as last-hour predictions (60.3% vs 59.8% correct outcomes; exact scores 9.7% vs 8.2%). Every lead-time bucket sat within a 4.2-point band (56.1%–60.3%), with no monotonic trend. Lock in your picks and enjoy the build-up.

Fans everywhere are equally accurate

Splitting predictors by account country produced a strikingly narrow band: every country with at least 100 predictors landed between 59.0% and 60.9% outcome accuracy (exact scores: 9.9%–11.1%). No national fan base in our sample out-predicted the rest by a margin that survives the sample sizes involved. Differences of a point or two across 7,000–25,000 predictions are within noise.

Inside the private leagues: lead changes and photo finishes

BeTeam predictions live inside private leagues — friends, offices, families — so we also measured how those competitions unfolded across the 350 leagues with at least 3 active players:

• 91.1% of leagues have had at least two different leaders; the median league saw its lead change hands 4 times in five weeks. In leagues with 21+ players, the median is 8 lead changes.
• The median gap between 1st and 2nd is 9 points going into the final — roughly three exact scores. 21.1% of leagues are within a single exact score (3 points), and 4.0% have joint leaders.
• 96.3% of leagues can still flip. With knockout multipliers, one exact prediction on the final is worth up to 48 points in classic scoring — more than the current top-2 gap in all but 13 of 350 leagues.

League standings are live values through the semi-finals (leagues conclude after the final); "can still flip" compares each league's top-2 gap against the maximum points one exact final prediction can earn under that league's own scoring settings.

Definitions

• Prediction — a member-entered final-score forecast, locked at kickoff. One per fan per match: if the same person predicted the same match in several leagues, only their earliest submission counts (194,812 of 205,229 raw predictions survive dedup).
• Correct outcome — predicted result class (home win / draw / away win) matches the recorded result. "Home" means the first-listed team; all 2026 World Cup venues are neutral.
• Exact score — both scores match the recorded final score exactly.
• Correct goal difference — predicted (home − away) equals actual; includes exact scores.
• Recorded result — the final score the app awards points against. Knockout matches decided on penalties are recorded as the after-extra-time draw (a shootout never changes the recorded score), matching BeTeam's published scoring rules.
• Random baseline — when a league member hasn't predicted by kickoff, the app files an automatic random score (each side uniform 0–3). These are excluded from all fan statistics and reported only as the baseline (n = 195,347 on the same matches).
• Crowd consensus — the outcome picked by the plurality of fans for a match; agreement = that plurality's share.
• AI prediction — a single score prediction per fixture from Google's Gemini (model gemini-2.5-flash) with Google Search grounding, generated and frozen before kickoff for the in-app stats screen. Stored for 19 knockout matches only (the feature launched July 9, 2026); compared against the same crowd and results as every other metric.
• Lead change — after any matchday, the member with the most points differs from the previous leader (ties retain the incumbent).

Methodology

Source. BeTeam's production PostgreSQL database, queried July 16, 2026 (data through July 15, 2026, 23:59 UTC). BeTeam is a free social prediction app for private groups; predictions carry no money and no odds.

Population. All member-entered, pre-kickoff score predictions on 2026 World Cup matches that (a) were real, automatically-scheduled fixtures (user-created games excluded), (b) finished with a publicly-reported final result — 102 matches: 72 group, 16 round-of-32, 8 round-of-16, 4 quarter-finals, 2 semi-finals. The bronze final and final had not been played at the data cut. Server-side validation rejects any prediction at or after kickoff; the dataset contains zero post-kickoff submissions.

Match results quoted here are the publicly-reported final scores of these matches — the same widely-published facts carried by every major results service. This report does not reproduce any data provider's proprietary statistics, compilations, or imagery; it publishes only public match results and BeTeam members' own predictions in aggregate.

Exclusions. Auto-generated random predictions (reported separately as the baseline); user-created manual games; cancelled or postponed fixtures; competitions other than the World Cup (each had under 20 fan predictions in the window — below our minimum publishable threshold); bonus-question answers (different mechanics from score predictions).

Deduplication. One prediction per (fan, match): earliest submission wins. 9,683 fan-match pairs appeared in more than one league; 4,066 of them — 2.1% of all deduplicated predictions — had conflicting scores across leagues.

Stages were assigned by kickoff timestamp and validated against the official match calendar (72/16/8/4/2). All timestamps are UTC.

Minimum samples. Country cells require ≥ 100 distinct predictors; scorelines below 1,000 predictions are aggregated; league statistics require ≥ 3 active players; every per-match figure rests on ≥ 1,709 distinct predictors.

AI comparison. The crowd-vs-AI section uses BeTeam's stored in-app AI predictions (Google Gemini gemini-2.5-flash, Google Search grounding, one per fixture, frozen before kickoff). Because that feature launched on July 9, 2026, AI predictions exist only for 19 knockout matches (Round of 32 → semi-finals) — the comparison is explicitly a knockout-round one, and the crowd side reuses the same deduplicated fan population defined above.

Reproducibility. Every figure is produced by a versioned, read-only SQL pipeline (14 queries + 2 derived computations) with 15 automated cross-checks (population sums, stage totals, share-sums, privacy floors, AI-comparison totals), all passing at publication. Statistical findings and interpretation are separated in the text; where we interpret (e.g. why fans under-predict draws), we say so.

Limitations

• One tournament. This is the 2026 World Cup through the semi-finals — 102 matches. Accuracy rates at club level, or at other tournaments, may differ. We will update the report with the final two matches after July 19, 2026.
• Self-selected population. BeTeam fans predict in private leagues with friends; they are engaged football fans, not a random sample of the public.
• No odds benchmark. We compare fans against a random baseline, not betting markets; "favorite" here means the crowd's own consensus, not bookmaker prices.
• AI comparison is small and knockout-only. The stored AI predictions cover 19 knockout matches (the feature launched mid-tournament), so "level with the AI" describes those 19 games, not the whole World Cup. One model (Gemini 2.5 Flash) with one prediction per match; a different model or prompt could score differently.
• Draw definition in knockouts. Four knockout matches level after extra time are counted as draws (as scored in-app); analyses that need unambiguous 90-minute results (draw share, goals averages) are restricted to the group stage.
• Timing metric. Submission time is when a fan first committed a prediction; pre-kickoff edits don't move it. League standings include points from auto-filled predictions (they are part of real leaderboards); league lead-change counts use match points only (bonus questions excluded).
• Correlation ≠ causation throughout: e.g. the calibration curve shows association between agreement and correctness, not that agreement causes anything.

Data & privacy notes

All statistics are aggregated and anonymized. No names, user identifiers, league names, invite codes, locations, or device data appear in this report or its downloadable files. The only user attribute used is account country, published solely for countries with at least 100 predictors. Per-match statistics aggregate 1,709+ fans each. Match names and results are public sporting facts. Questions: hello@beteamapp.com.

Download the data

Per-match crowd consensus (CSV, 102 rows) Accuracy by stage (CSV) Scoreline popularity & hit rates (CSV) Crowd vs AI, per knockout match (CSV, 19 rows)

Licensed CC BY 4.0 — free to use with attribution to BeTeam.

Data appendix: all 102 matches — crowd consensus vs result

Match	Stage	Result	Crowd pick	Agree	Crowd	Top score pick
Mexico vs South Africa	Group	2-0	Mexico	85.3%	✓	2-0
South Korea vs Czechia	Group	2-1	Draw	40.6%	✗	1-1
Canada vs Bosnia & Herzegovina	Group	1-1	Canada	63.4%	✗	2-1
USA vs Paraguay	Group	4-1	USA	68.3%	✓	2-1
Qatar vs Switzerland	Group	1-1	Switzerland	92.8%	✗	0-2
Brazil vs Morocco	Group	1-1	Brazil	76.5%	✗	2-1
Haiti vs Scotland	Group	0-1	Scotland	90.8%	✓	0-2
Australia vs Türkiye	Group	2-0	Türkiye	75.9%	✗	1-2
Germany vs Curaçao	Group	7-1	Germany	99.4%	✓	3-0
Netherlands vs Japan	Group	2-2	Netherlands	58.9%	✗	2-1
Côte d'Ivoire vs Ecuador	Group	1-0	Ecuador	39.3%	✗	1-1
Sweden vs Tunisia	Group	5-1	Sweden	79.4%	✓	2-0
Spain vs Cabo Verde	Group	0-0	Spain	99.4%	✗	3-0
Belgium vs Egypt	Group	1-1	Belgium	82.2%	✗	2-1
Saudi Arabia vs Uruguay	Group	1-1	Uruguay	82.5%	✗	0-2
Iran vs New Zealand	Group	2-2	Iran	53.7%	✗	1-1
France vs Senegal	Group	3-1	France	92.5%	✓	3-1
Iraq vs Norway	Group	1-4	Norway	88.7%	✓	0-2
Argentina vs Algeria	Group	3-0	Argentina	92.7%	✓	2-0
Austria vs Jordan	Group	3-1	Austria	77.9%	✓	2-0
Portugal vs DR Congo	Group	1-1	Portugal	98.8%	✗	3-0
England vs Croatia	Group	4-2	England	65.7%	✓	2-1
Ghana vs Panama	Group	1-0	Ghana	63.0%	✓	1-1
Uzbekistan vs Colombia	Group	1-3	Colombia	92.6%	✓	0-2
Czechia vs South Africa	Group	1-1	Czechia	70.7%	✗	2-1
Switzerland vs Bosnia & Herzegovina	Group	4-1	Switzerland	72.1%	✓	2-1
Canada vs Qatar	Group	6-0	Canada	77.2%	✓	2-0
Mexico vs South Korea	Group	1-0	Mexico	46.9%	✓	2-1
USA vs Australia	Group	2-0	USA	82.5%	✓	2-1
Scotland vs Morocco	Group	0-1	Morocco	83.4%	✓	0-2
Brazil vs Haiti	Group	3-0	Brazil	99.3%	✓	3-0
Türkiye vs Paraguay	Group	0-1	Türkiye	64.0%	✗	2-1
Netherlands vs Sweden	Group	5-1	Netherlands	68.6%	✓	2-1
Germany vs Côte d'Ivoire	Group	2-1	Germany	95.3%	✓	3-1
Ecuador vs Curaçao	Group	0-0	Ecuador	88.5%	✗	2-0
Tunisia vs Japan	Group	0-4	Japan	92.8%	✓	0-2
Spain vs Saudi Arabia	Group	4-0	Spain	95.0%	✓	3-0
Belgium vs Iran	Group	0-0	Belgium	91.2%	✗	2-0
Uruguay vs Cabo Verde	Group	2-2	Uruguay	87.4%	✗	2-0
New Zealand vs Egypt	Group	1-3	Egypt	72.6%	✓	1-2
Argentina vs Austria	Group	2-0	Argentina	96.0%	✓	3-1
France vs Iraq	Group	3-0	France	99.1%	✓	3-0
Norway vs Senegal	Group	3-2	Norway	56.8%	✓	2-1
Jordan vs Algeria	Group	1-2	Algeria	68.8%	✓	1-2
Portugal vs Uzbekistan	Group	5-0	Portugal	98.5%	✓	3-0
England vs Ghana	Group	0-0	England	97.4%	✗	3-1
Panama vs Croatia	Group	0-1	Croatia	95.1%	✓	0-2
Colombia vs DR Congo	Group	1-0	Colombia	80.5%	✓	2-1
Switzerland vs Canada	Group	2-1	Draw	38.1%	✗	1-1
Bosnia & Herzegovina vs Qatar	Group	3-1	Bosnia & Herzegovina	75.2%	✓	2-0
Scotland vs Brazil	Group	0-3	Brazil	97.3%	✓	0-2
Morocco vs Haiti	Group	4-2	Morocco	97.5%	✓	3-0
South Africa vs South Korea	Group	1-0	South Korea	83.3%	✗	0-2
Czechia vs Mexico	Group	0-3	Mexico	74.5%	✓	1-2
Ecuador vs Germany	Group	2-1	Germany	93.9%	✗	1-3
Curaçao vs Côte d'Ivoire	Group	0-2	Côte d'Ivoire	87.5%	✓	0-2
Tunisia vs Netherlands	Group	1-3	Netherlands	98.0%	✓	0-3
Japan vs Sweden	Group	1-1	Japan	62.5%	✗	2-1
Paraguay vs Australia	Group	0-0	Draw	38.2%	✓	1-1
Türkiye vs USA	Group	3-2	USA	78.4%	✗	1-2
Senegal vs Iraq	Group	5-0	Senegal	92.5%	✓	2-0
Norway vs France	Group	1-4	France	85.4%	✓	1-2
Uruguay vs Spain	Group	0-1	Spain	89.6%	✓	1-3
Cabo Verde vs Saudi Arabia	Group	0-0	Cabo Verde	49.4%	✗	1-1
New Zealand vs Belgium	Group	1-5	Belgium	91.8%	✓	0-2
Egypt vs Iran	Group	1-1	Egypt	67.3%	✗	2-1
Croatia vs Ghana	Group	2-1	Croatia	66.4%	✓	2-1
Panama vs England	Group	0-2	England	98.6%	✓	0-3
DR Congo vs Uzbekistan	Group	3-1	DR Congo	68.8%	✓	2-0
Colombia vs Portugal	Group	0-0	Portugal	75.2%	✗	1-2
Algeria vs Austria	Group	3-3	Austria	55.1%	✗	1-2
Jordan vs Argentina	Group	1-3	Argentina	99.5%	✓	0-3
South Africa vs Canada	R32	0-1	Canada	80.4%	✓	1-2
Brazil vs Japan	R32	2-1	Brazil	84.4%	✓	2-1
Germany vs Paraguay	R32	1-1	Germany	97.8%	✗	2-0
Netherlands vs Morocco	R32	1-1	Netherlands	55.0%	✗	2-1
Côte d'Ivoire vs Norway	R32	1-2	Norway	80.9%	✓	1-2
France vs Sweden	R32	3-0	France	97.9%	✓	3-1
Mexico vs Ecuador	R32	2-0	Mexico	70.4%	✓	2-1
England vs DR Congo	R32	2-1	England	96.5%	✓	2-0
Belgium vs Senegal	R32	3-2	Belgium	48.5%	✓	2-1
USA vs Bosnia & Herzegovina	R32	2-0	USA	88.8%	✓	2-0
Spain vs Austria	R32	3-0	Spain	95.8%	✓	2-0
Portugal vs Croatia	R32	2-1	Portugal	69.9%	✓	2-1
Switzerland vs Algeria	R32	2-0	Switzerland	71.8%	✓	2-1
Australia vs Egypt	R32	1-1	Egypt	68.8%	✗	1-2
Argentina vs Cabo Verde	R32	3-2	Argentina	95.2%	✓	3-0
Colombia vs Ghana	R32	1-0	Colombia	83.5%	✓	2-1
Canada vs Morocco	R16	0-3	Morocco	88.4%	✓	1-2
Paraguay vs France	R16	0-1	France	97.5%	✓	0-3
Brazil vs Norway	R16	1-2	Brazil	73.4%	✗	2-1
Mexico vs England	R16	2-3	England	54.8%	✓	1-2
Portugal vs Spain	R16	0-1	Spain	59.3%	✓	1-2
USA vs Belgium	R16	1-4	USA	44.5%	✗	2-1
Argentina vs Egypt	R16	3-2	Argentina	91.0%	✓	2-0
Switzerland vs Colombia	R16	0-0	Colombia	69.1%	✗	1-2
France vs Morocco	QF	2-0	France	84.2%	✓	2-1
Spain vs Belgium	QF	2-1	Spain	86.0%	✓	2-1
Norway vs England	QF	1-2	England	52.3%	✓	1-2
Argentina vs Switzerland	QF	3-1	Argentina	90.2%	✓	2-0
France vs Spain	SF	0-2	France	69.7%	✗	2-1
England vs Argentina	SF	1-2	England	43.2%	✗	2-1

How to cite this report

BeTeam Research (2026). How accurate are football fans? 194,812 World Cup 2026 predictions, analyzed. beteamapp.com. https://beteamapp.com/research/fan-prediction-accuracy/ (published July 16, 2026).

Journalists: figures may be quoted with attribution ("according to BeTeam, a social prediction app"). For the underlying aggregates or questions about methodology, email hello@beteamapp.com.

About this research

Produced by the BeTeam team from BeTeam's production database using the reproducible pipeline described above. BeTeam is a free iOS/Android app where friends, families, and coworkers run private football prediction leagues — members predict scores before kickoff, points are automatic, and a live leaderboard ranks the group. No deposits, no odds, no cash prizes; points are the only currency. Read more about BeTeam or see the World Cup 2026 engagement case study.

BeTeam is an independent app and is not affiliated with, endorsed by, sponsored by, or associated with FIFA or the FIFA World Cup. "World Cup" is used here only as a factual reference to identify the 2026 football tournament these predictions relate to.

Revision history

• July 16, 2026 — first publication, covering matches through the semi-finals (102 of 104). An update with the completed tournament is planned after the final on July 19, 2026.