Watch any football broadcast now and within minutes someone will mention "expected goals." A pundit says a team "deserved" three goals but only scored one, and the number they point to is xG. It sounds precise, almost scientific, but most viewers never learn what the number is actually built from or what it can and cannot tell them.

This matters because xG gets treated two very different ways: as gospel proof that a result was unfair, or as a gimmick that ignores what actually happened on the pitch. Neither view is right. xG is a tool for measuring shot quality, nothing more and nothing less, and understanding what goes into it is the difference between using it well and misusing it.

The simple idea behind expected goals

Think about buying a lottery ticket versus a scratch card that has a one-in-three chance of winning. Both are "chances to win," but they are not equally likely to pay off. Expected goals works on the same logic: not every shot on goal is an equal chance to score, so instead of counting shots as identical events, xG assigns each one a probability between 0 and 1 based on how likely a shot from that exact position and situation is to end up in the net.

A shot from six yards out, unmarked, might carry an xG of 0.65, meaning shots like it go in roughly 65 times out of 100. A speculative effort from 30 yards might carry an xG of 0.02, meaning it goes in about twice per 100 attempts. Add up all the shot values for a team across a match and you get a total xG score, something like 1.8 or 2.4, which represents the number of goals a team would be expected to score given the quality of chances it created, not the number it actually scored.

The core idea is separating "did they score" from "did they create good chances." A team can lose 1-0 while generating 2.3 xG, which tells you they created better chances than their opponent even though the scoreline says otherwise.

What data actually goes into an xG number

xG models are built by analyzing tens of thousands, sometimes millions, of historical shots and recording what happened to each one. The model then looks for patterns: which factors made a goal more or less likely. The main inputs typically include:

• Distance from goal: shots closer to the goal line score more often than shots from distance.
• Angle to goal: a shot from directly in front of the goal has a wider target than one from a tight angle near the touchline.
• Body part used: headers convert at a noticeably lower rate than shots struck with the foot.
• Type of assist or build-up: a through ball that puts a player one-on-one with the keeper produces a higher-value chance than a shot after a scrambled corner.
• Defensive pressure: whether defenders are between the shooter and the goal, and how many.
• Whether it's a set piece: penalties, free kicks and open-play shots are treated separately because their scoring rates differ sharply.

Different data providers, such as Opta, StatsBomb and Wyscout, build slightly different models with slightly different weightings, which is why you'll sometimes see two outlets report different xG totals for the same match. Neither is "wrong" exactly, they're just calibrated on different historical datasets and sometimes include different variables, like goalkeeper positioning.

Why two similar-looking shots can have very different xG values

Imagine two shots taken from the exact same spot, twelve yards out, dead center. On the surface they look identical. But if one striker is running onto the ball at full speed with the goalkeeper still repositioning, and the other striker is standing still while three defenders block the near post, the model will score these very differently, maybe 0.35 for the first and 0.08 for the second, because the surrounding context changes the true probability of scoring.

Here's a simple comparison table showing how much context can shift the number even when distance stays constant:

This is why simply counting "shots in the box" is a weaker measure than xG. Two players can each take ten shots from inside the box in a season, but if one player's shots are mostly tap-ins and the other's are mostly contested headers under pressure, their true scoring chances are nowhere close to equal, and xG is built specifically to capture that difference.

What xG is genuinely good at telling you

xG is strongest as a pattern-recognition tool over a sample of matches, not as a verdict on a single game. Over 10, 20 or 38 matches, a team's cumulative xG tends to correlate reasonably well with their actual goal output, because randomness in front of goal, like a striker's hot or cold streak, tends to smooth out over a larger sample.

It's also genuinely useful for spotting whether a team is generating quality chances or just racking up shot count. A team taking 18 shots a game but averaging only 0.9 xG is mostly taking low-value attempts from distance or tight angles. A team taking 9 shots but averaging 1.6 xG is getting into much better positions, even though the raw shot count looks worse. Analysts also use "xG difference," a team's own xG minus their opponent's, as a rough proxy for underlying team strength, often a better predictor of future results than the current league table, particularly early in a season when standings are still skewed by small sample size.

At the player level, comparing actual goals scored to expected goals over a season can flag finishing quality. A striker consistently scoring above their xG, say 18 goals from 14 xG across a season, is either an unusually good finisher or benefiting from favorable variance, and separating those two explanations usually requires looking at multiple seasons rather than one.

Where xG misleads people

The most common misuse is treating a single match's xG total as proof a result was "unfair." A team can post 2.1 xG and lose 2-0, but that single match includes randomness, goalkeeping brilliance, an offside call, a post that a shot rattled off, that no probability model captures. xG describes chance quality, not destiny, and a 90-minute sample is often too small for the numbers to average out.

xG also struggles with context that isn't captured in the shot-level data, including:

• Game state: a team already leading 3-0 may deliberately take fewer risks, lowering their own xG without meaning they created worse football.
• Goalkeeper quality: most public xG models score the shot, not the keeper, so a save from an elite goalkeeper looks identical in the data to a save from a backup.
• Rebounds and second phases: a blocked shot that leads to a tap-in can sometimes be undercounted or double counted depending on the provider's methodology.
• Time-wasting and fouls won: a team that's excellent at drawing fouls near the box, without ever shooting, generates zero xG despite controlling the game.

There's also a subtler trap: xG models are trained on historical average outcomes, so an exceptional finisher who consistently beats the model, or an exceptional playmaker who sets up chances the model underrates, will look "lucky" or "unlucky" for years even though what's really happening is individual skill the model wasn't built to detect.

Common mistakes

• Using one match's xG as the final word. A single game is too small a sample; look at 8 to 10 matches minimum before drawing conclusions about a team's underlying form.
• Ignoring which provider's model you're reading. Opta and StatsBomb numbers for the same match can differ by 0.3 or more, so compare like with like.
• Treating xG as a measure of "deserving to win." It measures shot quality, not effort, resilience or game management.
• Forgetting about game state. A team defending a lead will naturally show lower xG without playing worse football.
• Confusing shot count with shot quality. More shots does not automatically mean a better attack; check the average xG per shot too.

Using xG without over-relying on it

Next time you see an xG total on screen, treat it as one input among several, not a final scoreline. Ask how many matches the number is drawn from, whether the game state affected it, and whether a particular player's shots were mostly high-value or low-value chances. If you're tracking a team across a season, note both their xG and their actual goals scored every five matches or so; a growing gap in either direction is worth investigating rather than dismissing.

If you want to go deeper, look at "non-penalty xG" to strip out penalty conversions, which inflate totals for players who simply take spot kicks, and compare a striker's goals to their xG over at least two full seasons before deciding whether they're a reliably good finisher or riding a streak. Used this way, xG becomes what it was designed to be: a clearer lens on chance quality, not a replacement for watching the match itself.