What xG Gets Right and What It Doesn’t
Expected goals can tell you something the scoreline cannot. But it can also leave out parts of the game that matter. The trick is knowing what question xG is actually answering.
Brighton beat Arsenal 3–0. If that was the only information you had, you would probably imagine a straightforward story: Brighton created the better chances, Arsenal struggled to threaten, and three goals reflected the difference between the teams.
Then you look at the expected goals. Brighton: 1.33 xG. Arsenal: 1.58 xG.
So did Arsenal actually deserve to win? No. But Brighton scoring three goals does not mean xG failed either. That apparent contradiction is exactly why expected goals is useful — and why it is so often misunderstood.
First: what is xG actually measuring?
Expected goals looks at thousands of previous shots and estimates how often similar chances become goals. A 0.10 xG shot means similar attempts are scored about 10% of the time. It is a measure of chance difficulty, not a prediction of how many goals a team will score.
Every shot starts with a different probability.
The values below are illustrative, except the penalty value, which uses Opta's published 0.79 xG figure. Real shot values vary by model and context.
The value belongs to the chance before the outcome is known. A brilliant finish does not make the original opportunity easier; a bad miss does not make it harder. That distinction lets us separate the quality of a team’s attacking process from the quality of the final action.
Case Study 1: Brighton 3–0 Arsenal
Brighton scored through Pascal Groß, Charalampos Kostoulas and Chema Andrés. Arsenal finished with the higher xG, but Brighton were clinical — two of their goals came from low-probability finishes.
The total and the score told different stories.
This comparison uses verified team totals. A minute-by-minute race is not plotted because shot-level timestamps and coordinates were not available from the local provider.
What xG gets right
xG tells us that score and chance quality are not the same thing. Arsenal’s 1.58 xG says their shots carried slightly more combined scoring probability. Brighton’s three goals say their finishing on the day turned a smaller total into a decisive result.
That is not a conflict. One number describes the chances; the other records what actually happened.
Across a run of matches, this separation becomes even more useful. A team that repeatedly creates strong chances is usually building something more sustainable than a team relying on difficult finishes every week.
Where xG needs context
A single total does not tell you how the value was built. One penalty, several medium-quality chances and a long sequence of speculative shots can produce similar totals while describing very different performances.
Game state matters too. A team protecting a lead may give up territory and low-value attempts. A team chasing the game may accumulate shots without ever creating the chance it really needs.
Case Study 2: Forest 0–1 Coventry
Forest had 63% possession and took 24 shots, but only three were on target. Their total reached 1.15 xG. Coventry had ten shots, three on target and 0.67 xG — then scored the only goal.
The raw shot count makes Forest look dominant. The xG total qualifies that impression: 24 attempts produced only 1.15 xG, so the average shot was low value. Coventry created less, but converted one of the opportunities they did produce.
The totals can hide the shape of a match
xG is strongest when we ask what sits underneath it: Where were the shots taken? Did the total depend on a penalty? Were the chances created before or after the game changed? Did the goalkeeper face difficult finishes or comfortable ones?
Standard xG also begins with a shot. A dangerous cut-back that rolls centimetres beyond the striker records no xG because no shot occurred. The attack was dangerous, but a shot model cannot measure every valuable action before the shot. That is why the next layer of analysis needs concepts such as territory, pressing and transitions.
What xG tells me, and what I never ask it to do.
xG becomes much more useful once we stop treating it as an alternative scoreboard.
It distinguishes a 30-metre attempt from a close-range chance.
Across a run of matches it helps show whether a team is consistently creating dangerous shots.
It gives us a better way to discuss shot quality than treating every attempt equally.
2.0 xG does not mean a team should literally have scored two goals.
If an attack breaks down before the shot, standard xG records no chance.
Pre-shot xG values the chance before the finish. Post-shot models such as xGOT answer a different question.
Providers use different data and models, so the same chance can receive different values.
xG and xGOT are different questions
Pre-shot xG estimates the quality of the chance. Expected goals on target, or xGOT, looks at an on-target attempt after the player has struck it and considers where the ball was placed in the goalmouth. That helps evaluate finishing and goalkeeping without changing the original chance value.
xG and xGOT answer different questions.
The biggest mistake: treating xG as a verdict
If Brighton win 3–0, Brighton won 3–0. The xG total does not award Arsenal an alternative result. It helps explain that the result came from a combination of chance creation, finishing, goalkeeping and the way the match unfolded.
Use xG as evidence, not a verdict. When the number and your impression disagree, that is usually the beginning of the analysis, not the end.
Tifo Football explains why xG is powerful, where it can mislead and why the surrounding match context still matters.
Have you watched a match where xG completely disagreed with what you felt happened?
Share the match and explain what the number captured — or what it missed.
Why where you have the ball can matter more than how long you have it.
ThenWhat Actually Makes a Good Pressing Team?Pressing is organised pressure: triggers, traps, high regains and the risk behind the press.
What happens in the seconds after possession changes.
Data note: Match figures are the approved editorial figures supplied for this article. xG models differ between providers. Shot-level locations and timestamps were not available in the local match provider, so this version does not fabricate a shot map or minute-by-minute xG race.

