Expected Goals (xG) Explained: What It Measures and What It Misses

Expected goals is the most useful public football metric of the last decade and the most frequently misapplied. Both facts follow from the same source: it is simple to quote and harder to understand.

What xG actually is Every shot is assigned a probability of becoming a goal, based on how similar shots have finished historically. The inputs typically include distance from goal, angle, body part, type of assist, and whether the chance came from a set piece or open play.

A tap-in from two yards might carry 0.85 xG. A speculative effort from 30 yards might be 0.03. Add up every shot in a match and you get a team's xG for that game.

A team with 2.4 xG created chances that would, on average, produce about 2.4 goals. They may have scored none.

Why it beats the scoreline Finishing is unusually random over short periods. A team can dominate territory and chance quality and lose 1-0 to a deflection. Over a single match the scoreline is a poor description of what happened.

xG is more stable, which means it predicts future results better than past results do. A side consistently creating 2.0 xG and conceding 0.8 tends to start winning eventually, whatever the current table says.

Where it breaks This is the part usually skipped.

• Finisher quality is averaged away. An elite striker genuinely converts better than the model assumes. Treating that as unsustainable luck is a mistake.
• Models disagree. Different providers assign different values to the same shot. Comparing xG across sources is not valid.
• Game state distorts it. A team 2-0 up sits deep and concedes shots by choice. Their conceded xG rises while their actual risk does not.
• It ignores everything that is not a shot. A move broken up in the final third produces zero xG despite being dangerous.
• Small samples are noise. Single-match xG is nearly meaningless. Ten matches begins to say something.
• Penalties skew it. A penalty carries roughly 0.79 xG and arrives as a lump. One penalty distorts a single-game figure badly.

Using it sensibly Treat xG as one input describing chance creation and concession over a meaningful stretch — ten or more matches. Look for sustained gaps between xG and actual goals, then ask why. Is it finishing variance, or is it a specific striker, or a tactical pattern the model does not see?

The question is always "why is this gap here", never "this gap will close".

What we show Where our data covers enough finished matches, our match pages show recent scoring and conceding averages computed from real results in our database. We do not display an xG figure we have not independently verified, and we do not fill empty fields with estimates. Where a number is missing, the space stays empty — see our editorial policy for the reasoning.

Our football analysis desks are visible on the analysts page, where each one states its method openly, including which data it relies on.

xG describes what has happened. It does not predict what will happen, and no metric makes a result certain.