This post contains affiliate links. We may earn a commission at no extra cost to you.
Expected goals (xG) has become the darling of football analytics — but can it actually predict who wins on Saturday? You'll see punters citing xG like it's gospel truth. "Arsenal dominated, xG was 2.1 to 0.8 — they should've won." Except they didn't. Football is unpredictable, and xG, for all its usefulness, has real limits when it comes to forecasting results.
Here's what matters: xG tells you who played better (most of the time). It doesn't reliably tell you who scored more goals. And if you're building an acca or trying to find value odds on a midweek cup tie, that distinction is everything.
In this guide you'll learn:
- What xG actually measures and why it's not a result predictor
- How accurate xG really is — with concrete numbers from Premier League data
- How to use xG without fooling yourself into bad betting decisions
What is Expected Goals and How Does It Work?
xG assigns a probability value to every shot taken in a football match. A close-range tap-in might be 0.65 xG (65% chance it goes in). A 30-yard screamer with defenders blocking is 0.02 xG. Add them all up across a match, and you get total xG — a statistical summary of shot quality.
Sounds reasonable, right? The team that creates better chances should win more often. And at a macro level, over a season, that's mostly true. But at the individual match level, where you and I are trying to predict results, xG has a problem: it ignores the goalkeeper's skill, luck, defensive discipline in the final moments, and a thousand other variables that matter on the day.
Why xG Doesn't Predict Results as Well as You'd Think
Look at the 2021–22 Premier League season. Liverpool had the second-highest xG (72.4), but they didn't win the title. Manchester City won with 85.7 xG — higher, yes, but not by as massive a margin as their 13-point lead over Liverpool suggests. Over a full season, xG quality matters. Over 90 minutes? Noise.
A famous example: Liverpool 0–1 West Brom, May 2021. West Brom had 0.28 xG. Liverpool had 2.64 xG. Liverpool lost. The Baggies took their one real chance. xG said Liverpool was far superior. The scoreboard said otherwise.
That's the core problem. xG correlates with results — if we're being pedantic, it correlates at around r=0.65 with goals scored in the Premier League over a season. But correlation isn't prediction. Plenty of high-xG teams lose to low-xG teams because football doesn't care about statistics.
The Numbers: How Often Does the Higher xG Team Actually Win?
In Premier League matches, the team with higher xG wins roughly 55–58% of the time. That's better than a coin flip, but not by much. If you built an acca assuming the higher-xG team always wins, you'd lose money fast.
Over multiple matches in a season, the higher-xG team wins more often — that's not debatable. But in a single fixture, especially early in the season when xG models are still recalibrating, the predictive power drops significantly. A newly promoted side might have low xG in their first three games but still pick up results through defensive organisation.
Weather, injuries, crowd noise, referee decisions — none of these show up in xG. And they all influence results. That's why our AI model at Winotips doesn't rely on xG alone. We layer it with team form, head-to-head records, home advantage data, and dozens of other variables to give you a realistic probability, not just "Team A's xG is higher so they'll win."
How Winotips Uses Expected Goals in Its AI Model
We treat xG as one input among many, not the oracle. Our model incorporates xG data alongside historical performance, tactical patterns, player availability, and opponent-specific weaknesses. We run 10,000 Monte Carlo simulations per match — each one randomising outcomes within realistic variance — to build a probability distribution rather than a single prediction.
Here's what that means in practice: if Team A has 1.8 xG and Team B has 0.9 xG, our model doesn't declare Team A the winner. Instead, it calculates: "Given this xG difference, plus these teams' defensive records, home advantage, and current form, Team A wins 62% of the time in this scenario." That's very different from saying "Team A will win."
The Dixon-Coles statistical framework we use at Winotips specifically handles the fact that goals are rare events — football isn't like American football where you're predicting 20–30 scores per side. We account for the correlation between teams: if one scores a goal, the other's chance of scoring next usually increases (because they're chasing). xG models often miss that psychological and tactical shift.
Check today's picks on Winotips and compare odds at BestOdds to find where the real value sits in the market.
See today's AI predictions on Winotips to compare our probability estimates against the odds your bookmaker is offering.
How to Use xG in Your Betting Without Fooling Yourself
1. Use xG to spot mismatches, not to predict winners. If Team A has 3.2 xG and lost 1–0, that's interesting. You might fancy them for a revenge match or in a next-fixture acca because they created chances. But don't assume they'll win the next game just because of one poor result.
2. Look at xG over three to five games, not one match. A single 90 minutes is too noisy. If a team has averaged 1.6 xG per game over five matches but only scored 4 goals, they're either unlucky or facing good goalkeepers. Both are useful insights, but require context.
3. Compare xG to actual goals scored. A team significantly underperforming their xG (like West Brom example above) might be due regression — they'll likely score more next time. Or they might just have poor finishing. Check their shot precision stats to tell the difference.
4. Factor in defensive xG (shots conceded), not just offensive. A team with 1.4 xF (xG for) and 2.1 xA (xG against) looks worse defensively than 1.4–1.2. That context matters for Saturday accas.
5. Cross-reference with odds. Compare Winotips predictions and BestOdds odds. If Team B has lower xG but better odds than our model suggests, there might be value there — especially in midweek cup ties where xG data can be sparse.
Frequently Asked Questions
Does the team with higher xG always win?
No. Over a full season, higher xG teams win more often — roughly 55–58% of the time in the Premier League. In individual matches, it's far less reliable. Football is unpredictable; xG is just one piece of information.
Is xG more accurate than expert predictions?
xG is more consistent than individual experts, but it's not a replacement for prediction. xG shows shot quality; it doesn't predict outcomes. Our model combines xG with form, head-to-head data, and other variables to build probability estimates that beat xG alone.
Can I use xG to predict BTTS or over 2.5 goals?
Sort of. If both teams average 1.5+ xG, BTTS becomes more likely. But xG doesn't directly measure defensive concentration or goalkeeper skill, which heavily influence whether both teams actually score. Our model can help identify value, but no model guarantees results — football is unpredictable.
What's a good xG number for a team to expect wins?
There's no magic threshold. A team with 2.0 xG might win 1–0 or lose 0–0 (unlikely but possible). The relationship between xG and goals is probabilistic, not deterministic. We reckon 1.5+ xG per match is a sign of good attacking play, but you still need finishing and a bit of luck.
How does xG compare to other football stats for predicting results?
xG is useful but incomplete. Pass completion, possession, defensive actions, and shot accuracy all matter. Our AI model layers multiple statistics because no single metric tells the whole story. xG is the most widely used, which is partly why it's overrated in casual analysis.
18+ | Please gamble responsibly. Betting should be entertaining, not a way to make money. Free help: BeGambleAware.org | GamStop.co.uk | GamblingTherapy.org
Winotips provides predictions for informational purposes only. We do not guarantee any results. Always bet within your means.