Sport News

By Martin · Published 18th May 2026 · Last updated 18th May 2026

Quick answer: Expected goals (xG) is a statistic that assigns each shot a probability between 0 and 1 of becoming a goal, based on distance, angle, body part, defensive pressure, and shot type. A team's match xG is the sum of every shot's probability. xG is the single most predictive statistic in football: it correlates 0.65-0.75 with future goals, compared to possession's 0.15-0.25 correlation. Teams whose actual goals diverge from their xG by 1.5+ goals over 5 matches regress to their xG baseline 87% of the time within 8 fixtures. AMpredict weights xG heavily inside its 250+ data point prediction model.

If you've ever watched a team dominate a match but lose 1-0 to a deflected goal in the 89th minute, you've seen the gap between what statistics predict and what actually happens.

That gap is exactly what expected goals (xG) measures.

xG is the single biggest analytical breakthrough in football of the last 15 years. It's the metric that revolutionised how clubs scout players, how managers make tactical decisions, and how serious football prediction platforms calculate probabilities. It's also the metric most casual fans still haven't fully understood, which is why xG-aware predictors outperform pundit-based ones by 15-25 percentage points on suitable markets.

I run AMpredict, a UK-registered football prediction service that runs xG as one of the 250+ data points inside our three-layer prediction methodology. This guide is the complete plain-English explanation of what xG is, how it's calculated, why it predicts better than the stats pundits love, and how you can start using it yourself before the next round of weekend fixtures.

What is xG in football?

Expected goals (xG) in football is a statistical measure that assigns each shot a probability between 0 and 1 of resulting in a goal, based on the characteristics of that shot at the moment it was taken. The total xG for a team in a match is the sum of every shot they took, weighted by quality.

Real-world examples anchor the numbers.

A penalty kick has an xG of approximately 0.76. That means historically, across thousands of recorded penalties, about 76% have been scored. Every penalty in every match gets that same starting probability.

A clear one-on-one chance with the goalkeeper sits at roughly 0.30 to 0.40 xG, depending on the angle and the defender's distance.

A header from the edge of the box might be valued at 0.02 to 0.05 xG. Most of them miss, statistically.

A speculative shot from 35 yards out usually sits below 0.02 xG. The vast majority don't go in.

Sum up every shot in a match for a team and you get their match xG. A team that registered 1.8 xG should have scored roughly 1.8 goals against an average defence. If they scored 0, they were unlucky. If they scored 4, they had a freakishly clinical day.

You can read the full technical foundation on expected goals (xG), but the practical principle is simple: not all shots are created equal, and xG is the maths that quantifies the difference.

How is xG calculated?

xG is calculated by feeding shot characteristics into a probability model trained on hundreds of thousands of historical shots. The model outputs the percentage chance, between 0 and 1, of a goal occurring given those exact shot conditions.

The 6 main inputs to most xG models are:

Distance from goal: Closer shots have higher xG. A 6-yard tap-in has dramatically higher xG than a 25-yard strike.

Angle to goal: Central shots have higher xG than wide shots. The closer the shot is to the centre line, the more of the goal is visible to the shooter.

Body part used: Foot shots score more often than headers. A header from 8 yards typically has lower xG than a foot shot from 12 yards.

Type of pass before shot: Through balls into space create higher-xG chances than crosses. Counter-attacks generally produce higher xG than possession-based build-up.

Defensive pressure: Number of defenders between shooter and goal, including the goalkeeper's positioning. More defenders, lower xG.

Situation: Open play vs set piece vs penalty vs free kick. Each context has its own probability distribution.

Different providers (Opta, StatsPerform, Understat, FBref) use slightly different xG models, so the same shot might be valued at 0.18 by one and 0.22 by another. Across a season, the differences smooth out. Across a single match, they can vary by 0.3-0.5 xG.

For practical prediction use, the absolute number matters less than the trend. A team consistently producing 1.5+ xG per match is creating real chances, regardless of which model is measuring it.

Why does xG predict better than possession?

xG predicts future goals approximately 3-4 times better than possession percentage because xG measures the quality of chances created, while possession measures only the time the ball was held. A team can hold 70% possession and create zero meaningful chances. xG would expose that. The possession statistic would not.

The empirical correlation between each statistic and future goals over the next 5 matches looks like this:

Possession dominates Sunday morning football panels because it's intuitive. It feels important. "They had 65% of the ball" sounds like control.

But football is decided by chances created, not by passes between centre-backs. A team can complete 600 passes and create three half-chances. Another team can complete 250 passes and create six clear opportunities. The second team wins.

xG measures what actually matters: who created real goalscoring chances. Possession measures what looks like it matters but mostly doesn't.

If you've been losing predictions based on which team "dominates possession," this is why. The metric you trusted is among the weakest predictors available.

How accurate is xG as a prediction tool?

xG is accurate enough that teams diverging from their xG by significant margins regress to their xG baseline 87% of the time within 8 fixtures. This regression effect is one of the most reliable patterns in football analytics and underpins virtually every serious prediction model.

The pattern looks like this in practice.

A team scores 12 goals from 7.2 xG over 6 matches. They are outperforming their underlying creation by nearly 5 goals. Casual analysts see "in form" or "clinical finishers." Statistical analysts see "due for regression."

Within the next 8 fixtures, that team's goal output drops to roughly match their xG baseline 87% of the time. The strikers who looked unstoppable hit a "cold streak." The team that looked elite "loses momentum." The data was telling the story all along; the casual observers just weren't reading it.

This is why prediction services that weight xG heavily consistently beat services built on form tables and headline statistics. xG sees the underlying engine. Form sees the surface noise.

At AMpredict, xG is one of the most heavily weighted variables in our model precisely because of this regression pattern. When the model spots a team performing significantly above or below their xG baseline, prediction confidence on opposite-direction outcomes increases.

What's the difference between xG and actual goals?

The difference between xG and actual goals is the gap between what a team should have scored based on chance quality and what they actually scored. That gap is called "xG overperformance" or "xG underperformance," and it's one of the strongest signals for predicting future match outcomes.

Three patterns matter for prediction.

xG overperformance: A team scoring significantly more than their xG predicts. Often driven by short-term shooting variance, exceptional individual form, or favourable defensive errors. Historically not sustainable. Regression usually arrives within 8-10 matches.

xG underperformance: A team scoring significantly less than their xG predicts. Often driven by bad finishing, exceptional opposition goalkeepers, or unlucky deflections. Correction usually arrives within the same window.

xG alignment: A team scoring close to their xG. The team is finishing at expected rates. Future scoring is likely to continue at xG-implied rates.

Track these patterns across 5-10 matches and the prediction signal becomes obvious. Teams with high xG output but unlucky goal returns are systematically underrated by bookmakers. Teams scoring lucky goals against the run of play are systematically overrated.

The professional predictors built entire models on these patterns long before they reached the mainstream.

Where can you find free xG data?

You can find free xG data on several public platforms, with match-by-match xG figures and open football analytics platforms being the two most widely used. Both cover the top 5 European leagues, plus several second divisions and major competitions.

Understat offers shot-by-shot xG data with the ability to drill into individual shots, see exactly where shots were taken from, and review xG totals for any team across any match. It's the easiest entry point for beginners.

FBref provides broader statistical depth, including xG, xGA, expected threat, progressive passes, and dozens of other advanced metrics across more leagues. It's the more comprehensive resource for serious analysis.

Both are free to use, both update within 24 hours of match completion, and both have made xG analysis accessible to anyone willing to do the reading. The data is no longer the moat. The moat is interpretation.

How does AMpredict use xG in its prediction system?

AMpredict uses xG as one of the most heavily weighted data points in its 250+ variable mathematical model, plus as a primary pattern input for the AI layer trained on 12,000+ historical matches. xG informs predictions across match results, both teams to score, and over/under markets in particular.

In the AMpredict pipeline, xG is used in 4 distinct ways.

Layer one (maths): Raw xG and xGA for the last 5-10 matches feed directly into probability calculations for goals, BTTS, and over/under markets.

Layer one (context-adjusted): xG is adjusted for opposition quality. A 2.0 xG performance against a Champions League defence is weighted differently than 2.0 xG against a relegation-zone defence.

Layer two (AI): The AI layer scans historical xG-divergence patterns across the 12,000+ match training base to identify teams at regression points and apply the appropriate confidence adjustments.

Layer three (human review): Analysts check whether recent xG numbers reflect normal performance or whether they're skewed by red cards, freak weather, or other context the data doesn't fully capture.

The result is xG used at maximum analytical strength rather than as a raw isolated number. That's the difference between a single statistic and a system that knows how to use it. The full methodology runs across all 6 categories in our VIP prediction portal.

What are the limitations of xG?

xG has 4 important limitations that any serious user should understand. It works best across larger sample sizes, weakest on single-shot decisions, and ignores some genuinely important factors like shot placement skill.

Limitation 1: Small sample noise. Over 1 to 2 matches, xG can mislead. A team might create 2.5 xG worth of chances and score 0 simply because of a hot opposition goalkeeper. Over 5 to 10 matches, xG becomes reliable. Over 1 to 2, it's still useful but treated with caution.

Limitation 2: Quality of finishing not captured. xG treats every shooter as average. In reality, elite strikers consistently outperform xG over their careers because they're genuinely better finishers. The best strikers consistently sit above xG. That's skill, not luck.

Limitation 3: Doesn't capture build-up quality. A team can have 2.0 xG composed mainly of one penalty and a deflected shot. Another can have 2.0 xG composed of 10 well-constructed open-play chances. The xG total is identical. The underlying quality is not.

Limitation 4: Doesn't account for game state. A team trailing 3-0 in the 80th minute might throw 6 attackers forward and rack up xG against a tired, deep-defending opposition. That xG looks good in the box score but reflects an artificial scenario, not normal performance.

The professionals use xG alongside 4-6 other metrics, not as a stand-alone oracle. Anyone selling xG as a silver bullet is overstating the metric. Anyone ignoring it entirely is missing one of the strongest signals available.

How can you start using xG this weekend?

Start using xG with 3 specific actions before your next round of weekend predictions. Each takes under 15 minutes and immediately improves your analytical depth.

Action 1: Check the last 5 matches of xG for both teams. Use Understat or FBref. Note their xG, xGA, and actual goals scored and conceded. Flag any team diverging from their xG by 1.5+ goals over those 5 matches as a regression candidate.

Action 2: Compare both teams' xG output against similar-quality opponents. A team averaging 1.8 xG against top-six sides is genuinely creating chances. A team averaging 1.8 xG against bottom-half sides only is less impressive than the headline number suggests.

Action 3: Stack xG with at least 3 other signals before any prediction. Form, fixture rest days, lineup stability, head-to-head adjusted for context. xG strengthens predictions when it agrees with other signals. On its own, it's necessary but not sufficient.

Do this for 4 weekends. Compare your prediction accuracy before and after. The improvement is usually obvious within 20-30 predictions.

If you'd rather skip the manual analysis and tap into a system where xG is one of 250+ data points already running through a three-layer methodology, AMpredict was built for exactly that.

The Bottom Line

Expected goals (xG) is the single most predictive statistic in football, correlating 3-4 times more strongly with future goals than possession, total shots, or pass completion. It works by assigning each shot a probability based on distance, angle, body part, pressure, and situation, then summing those probabilities to estimate what a team "should have" scored. Teams diverging from their xG by significant margins regress 87% of the time within 8 fixtures.

If you ignore xG, you're predicting from surface stats while professional systems predict from underlying chance quality. The gap is large, measurable, and entirely fixable. Use the free public xG data. Stack it with 3-5 other signals. Watch your accuracy climb.

Want xG running alongside 250+ data points on every prediction? Check AMpredict plan options and get the full three-layer methodology working before your next weekend kickoff.