The problem with xG is people sometimes try to make it out to represent what the score should have been, when all it really represents is what on average would the expected outcome of goalscoring opportunities in the game normally be, without considering how well the players actually did in those situations.
The reason it doesn't represent what the score should have been is because the actual goals scored is influenced by the performances of the players. So for example if Ramsdale makes a couple of great saves against Liverpool, that doesn't mean Arsenal were lucky not not concede or that Liverpool were unlucky not to score. It means one of the Arsenal team executed his job very well. Or conversely if Rosenthal misses an open goal then his team don't deserve that goal, or if a keeper does a Karius and just lets a weak shot squirm straight through him, it means Xg and actual goals are misaligned but based on the performances of the players on the pitch, the actual goals reflects the situation.
So in essence xG is probably a decent metric of the quantity and quality of chances teams are creating, but shouldn't be used to try and argue what the outcome of the match should have been.