Expected goals, or xG, measures the likelihood that a single chance is scored, based on how similar past shots turned out. That is how Opta defines it, and it answers what xG measures in football: the quality of an individual shot at the moment it is taken, not the quality of a team or a match.
Opta says its model uses shot location plus dozens of on-field contextual factors. A match total is the sum of individual shot probabilities, while goals are discrete outcomes, which is why a side with the higher xG can still lose. Providers build their models differently, so figures from one are not directly comparable with another’s.
This explainer is a concept piece, not a match analysis. It sets out what the model sees, what freeze-frame data from Hudl Statsbomb adds, how xGOT differs from xG, and where the number stops. The provider descriptions are vendor material and are presented here as definitions, not independent evaluation.
Quick answer: Opta defines xG as the likelihood that a chance is scored, based on similar past shots, using shot location plus dozens of on-field contextual factors at the moment of the shot. xGOT applies to on-target shots and adds shot execution quality to chance quality. xG estimates chance quality and is not a measure of the whole match or a guaranteed result predictor.
What the model sees at the moment of the shot
Before a shot is taken, the model has nothing to score. Its value is fixed only at the moment the ball is struck, and that value comes from a comparison with history rather than from what the match looked like.
Opta, the data arm of Stats Perform, defines expected goals (xG) as the likelihood that a chance is scored, based on similar past shots (Opta Football Stats Definitions, published by The Analyst). The logic is a lookup: take a shot, find comparable attempts from the past, and ask how often those went in. The resulting figure is a probability between zero and one, attached to a single attempt.
The inputs matter more than the label. According to Stats Perform’s own description of its models, the Opta xG model uses shot location plus dozens of on-field contextual factors at the moment of the shot. The published description does not itemise those factors here, so this article does not list or count them. Location is the anchor, and the context refines it.
Two caveats follow for anyone reading the figures closely. First, both the definition and the model description come from the vendor that builds and sells the product. They tell us what Opta says its model does, not how it performs against rivals in an independent test. Second, the phrase “at the moment of the shot” sets the boundary: the model values the attempt as it stands when taken. It does not value the passing sequence that preceded it, or the shape of either team before the ball arrived.
In our view, that boundary is the most useful thing to hold onto. xG answers one narrow question about one attempt, and everything else is a separate matter.
How chance creation feeds into the number
Match xG is a sum, and that shapes how it should be read. According to a 2026 paper in ScienceDirect, a team’s xG for a match is the total of the individual shot probabilities, whereas goals are discrete outcomes: a shot is scored or it is not. A single attempt valued at a given probability can only ever produce zero goals or one. The running total across a match is an expectation built from many such attempts, not a prediction of the scoreline.
That is why a team can finish a match with a higher xG total and still lose. The total describes the quality and number of the shots taken. It does not guarantee that those shots convert, and it says nothing about what the other side did with theirs. In our view, this is the plainest reason to treat xG as a probabilistic estimate of chance quality, not as a measure of the whole match or a guaranteed result predictor.
Where the credit sits

Opta defines xG as the likelihood that a chance is scored, based on similar past shots. The unit of valuation is therefore the shot. The model attaches a probability to each attempt, and the passes, carries and movements that led to it do not receive a figure of their own in that definition.
Whether any provider’s model credits build-up play in some indirect way is not established by the sources supplied for this piece, and we do not claim it either way. It would need to be checked against provider documentation before anyone states it as fact. What the sources do support is narrower: the stored definition values the chance at the point of the shot.
No team or match xG figures are given here. This is a conceptual explanation, and any number quoted without its provider, competition and date would mislead more than it informed.
For readers comparing teams, the practical consequence is modest but important. A season-long xG total tells you about the shots a side took and allowed, valued one by one. If you want to judge build-up, you need other measures and other evidence.
Pressure and goalkeeper position: what freeze frames add
Can xG register the pressure a shooter is under? Early public models often could not. According to a Hudl Statsbomb post on X dated 3 February 2022, coaches working with older data would object that expected goals models “can’t tell how much pressure the shooter is under” and “don’t know where the goalkeeper is”. The same post says the provider responded by including defender, attacker and goalkeeper positions at the moment of the shot, known as freeze frames. This is the provider’s own account of its history, not an independent audit.
A freeze frame is a snapshot of the players visible when a shot is taken. In principle it lets a model separate two shots from the same spot: one with a defender closing in and a goalkeeper set, and one taken in space with the goalkeeper out of position.
What the provider says the model now includes
Hudl’s guide to football event data, published on 9 October 2026, is reported to state that Hudl Statsbomb xG includes goalkeeper position, all attackers and defenders in frame, and shot impact height. We have not verified that page in full, so treat this list as the vendor’s description of its own product, not as confirmed fact. The factors are not enumerated here beyond what the sources state.
Two limits follow. First, a freeze frame records where players stand, not what they intend, so it cannot show how a defender’s pressing contributed to that moment. Second, the sources supplied do not say how much these inputs change any given value, and no figure is quoted here. In our view, freeze-frame xG describes the shooting situation more fully, but it remains a measure of the shot, not of the press that preceded it.
Chance quality versus shot execution on target
Expected goals and expected goals on target (xGOT) answer two different questions. xG asks how likely a chance was to be scored. xGOT asks how likely a shot that reached the target was to be scored, once the way it was struck is taken into account.
Opta defines xG as the likelihood that a chance is scored, based on similar past shots, according to its published stats definitions. The value is fixed by the situation at the moment of the shot. Whether the ball then flies into the top corner or straight at the goalkeeper has no bearing on it.
xGOT changes that. According to Stats Perform, Opta’s xGOT model credits on-target shots for shot execution quality as well as chance quality. The two ideas are layered: the model starts from the quality of the opportunity and then adds how well the shot was taken.
Two consequences follow from the definitions. First, xGOT applies only to shots on target, so a shot that misses the frame has an xG value but no xGOT value. Second, the gap between the two measures on a single shot reflects execution, not the chance itself. A shot that earns a higher xGOT than its xG was, by the model’s logic, better placed or struck than the average attempt from that situation.
The sources supplied do not give event sequences or worked examples, so none is offered here. The numbers they do give are definitions, not independent tests, and both come from the provider that sells the model. In our view, xGOT is best read as a description of finishing and shot-stopping, while xG remains the description of the opportunity.
Why xG values differ between providers
Two providers can publish different xG for the same shot because each builds its own model. In our view, that means xG figures from one provider are not directly comparable with another’s. That observation rests on a 2026 ScienceDirect paper on probabilistic reformulations of expected goals, which notes that models differ by provider. It does not publish a side-by-side comparison of any single shot, so none is offered here.
Different inputs, different estimates
Opta, part of Stats Perform, defines xG as the likelihood that a chance is scored, based on similar past shots. Its model uses shot location plus dozens of on-field contextual factors at the moment of the shot, according to Stats Perform. The full list of those factors has not been published in the material available to us, so we do not name them.
Hudl Statsbomb describes a different set of inputs. According to Hudl’s guide to football event data, its xG includes goalkeeper position, all attackers and defenders in frame, and shot impact height. This is the freeze-frame approach. Hudl Statsbomb said in a 2022 post on X that it added defender, attacker and goalkeeper positions after coaches objected that older models could not see how much pressure the shooter was under or where the goalkeeper stood.
Both descriptions come from the companies that sell the models. They are definitions of what each model includes, not independent tests of which is more accurate.
What the difference means for readers
A model that reads defender and goalkeeper positions is working from different information than one built on location and contextual factors. It would therefore be unsurprising, in our view, for the two to assign different values to the same chance. We cannot show the size of any gap, because the sources supplied give no paired figures.
The practical rule is to name the provider and keep comparisons within it. Set one provider’s season totals against another’s and the difference may come from the model rather than the football.
The publication's view: a probability, not a verdict on the match
EDITORIAL. In our view, xG is best read as a statement about the quality of chances, not a ruling on who deserved to win. The figure for a match is a sum of individual shot probabilities, according to a 2026 ScienceDirect paper on probabilistic reformulations of xG. Goals, by contrast, are discrete outcomes: a shot either goes in or it does not.
That gap explains why a side with the higher xG total can still lose. A chance rated as likely to be scored is not a goal, and a low-value effort that goes in is still a goal. The model describes how often shots of that kind have been scored, not what will happen on a given afternoon.
The limits follow from that. xG is a probabilistic estimate of chance quality, not a measure of the whole match and not a guaranteed predictor of the result. Opta describes its own figure as the likelihood that a chance is scored, based on similar past shots, so the number speaks to shots and nothing beyond them.
A sensible reading, then, is to treat xG as one piece of evidence alongside the score. Name the provider, keep comparisons within one model, and use the figure to ask whether a result matched the chances created, not to overrule it.
Sources
- pmc.ncbi.nlm.nih.gov · Retrieved October 25, 2026
- Bayes-xG: player and position correction on expected goals (xG) using Bayesian hierarchical approach · Retrieved October 25, 2026
- The Nuances of Expected Goals (xG): Understanding Football’s Most Misunderstood and misused… · Retrieved October 13, 2026
- Beyond Expected Goals: A Probabilistic Framework for Shot Occurrences in Soccer · Retrieved October 13, 2026
- What Does Expected Goals (xG) Really Tell Us? · Retrieved October 13, 2026
- What Is Expected Goals (xG)? · Retrieved October 13, 2026
- Biases in Expected Goals Models Confound Finishing Ability · Retrieved October 13, 2026
- Expected goals, expected assists, pressures, carries, high turnovers and more · Retrieved October 13, 2026
- What are Expected Goals and what do they really tell us? – Barça Innovation Hub · Retrieved October 13, 2026
- Expected Goals (xG) Explained: Best Guide to 6 Key Inputs · Retrieved October 13, 2026
- xG (Expected Goals) for Dummies: What It Is, Why It Matters, and How It’s Changing Soccer · Retrieved October 13, 2026
- Limitations of expected goals · Retrieved October 13, 2026
- A probabilistic and dynamic reformulation of expected goals (xG): Methodological advances and modelling perspectives – ScienceDirect · Retrieved October 13, 2026
- What Is xG in Soccer? Expected Goals Explained Simply — DataField.Dev Blog · Retrieved October 13, 2026
- Opta Football Stats Definitions · Retrieved October 13, 2026
- Expected Goals (xG): The Football Metric Changing Analysis, Betting, and Fan Engagement · Retrieved October 13, 2026
- Enhancing Expected Goals on Target: Deeper Insights Into Shot Execution, Goalkeeper Performance, and a Dedicated Model for Women's Football · Retrieved October 13, 2026
- What Are Expected Goals on Target (xGOT)? · Retrieved October 13, 2026
- Post-Shot Expected Goals (PSxG): What is it and why does it matter? · Retrieved October 13, 2026
- Expected Goals (xG) Explained: Football’s Key Metric Simplified – StatsUltra · Retrieved October 13, 2026
- Hudl Statsbomb on X: "Freeze Frames on Shots When using old data, coaches would point out: "But expected goals models can't tell how much pressure the shooter is under, and they don't know where the goalkeeper is" So we included Defender, Attacker, and Go · Retrieved October 13, 2026
- The Reporter's Guide to Football Event Data · Retrieved October 13, 2026
- Enhancing xG models with freeze frame data · Retrieved October 13, 2026
- GitHub – ShrxshCode/xG · GitHub · Retrieved October 13, 2026
- Hudl Statsbomb – The World’s Most Advanced Soccer Data · Retrieved October 13, 2026
- Upgrading Expected Goals • Statsbomb Blog Archive · Retrieved October 13, 2026
- (PDF) Improving expected Goals (xG) models: methods, results, and novel extensions · Retrieved October 13, 2026
- Visualizing Soccer with StatsBomb Data and R, Part 1: Simple xG and Pass Partner Plots! · Retrieved October 13, 2026
- www.frontiersin.org · Retrieved October 13, 2026
- www.businesswire.com · Retrieved October 25, 2026
- Founded Year · Retrieved October 25, 2026
- HustleRate: Measuring the Relentless Press of Premier League Forwards · Retrieved October 25, 2026
- Forced Turnover: Evaluating Pressing Effectiveness in Soccer · Retrieved October 25, 2026
- Central European Journal of Sport Sciences and Medicine (Jan 2021) · Retrieved October 25, 2026
- IDEAS home Printed from https://ideas.repec.org/a/jsf/intjsf/v7y2012i3p213-231.html · Retrieved October 25, 2026
- arxiv.org · Retrieved October 25, 2026
- www.businesswire.com · Retrieved October 25, 2026
- soccer penalty shootouts · Retrieved October 25, 2026
- StatsBomb Data Case Studies: Freeze Frames And Defender Locations • Statsbomb Blog Archive · Retrieved October 25, 2026
- Expected Goals (xG) Explained: What It Actually Measures · Retrieved October 25, 2026
- apps.apple.com · Retrieved October 25, 2026
- A Quick Timeout · Retrieved October 25, 2026
- An examination of expected goals and shot efficiency in soccer · Retrieved October 25, 2026
- Integrating Dynamic Defensive Geometry and Match-State … · Retrieved October 25, 2026
- The Interpretive Limits of Expected Goals: Model Dependence, Realization Variability, and Decision-Making Under Uncertainty · Retrieved October 25, 2026
- Frontiers · Retrieved October 25, 2026
- (PDF) Beyond Expected Goals: A Probabilistic Framework for Shot Occurrences in Soccer · Retrieved October 25, 2026
- Bayes-xG: Player and Position Correction on Expected Goals (xG) using Bayesian Hierarchical Approach · Retrieved October 25, 2026
- Expected goals · Retrieved October 13, 2026
- Penalty shot (ice hockey) · Retrieved October 25, 2026








Leave a Reply