"AI" is a label, not a method, and it is not the word we use for what we run. The same two letters cover a rating system from the 1960s and a neural network trained last month.
Nearly every published forecast comes from a rating system, a general machine-learning method such as gradient-boosted trees, or a goals model. We run a goals model: attack and defence strength per team on top of the league scoring rate and a fitted home-advantage term, weighted to recent form and blended with expected goals, turned into a Dixon-Coles-corrected Poisson distribution over scorelines, with every market read off that one grid and then anchored partway to the bookmakers' price with the margin removed.
It can treat every match the same way, stay calibrated, and be corrected when the grading shows a systematic error. It cannot remove the randomness from football, promise an outcome, or see motivation, team news and a manager appointed on Thursday. The bookmakers' closing price has already absorbed most of what a model does not know.
A selection cannot be checked; a probability can. Every number we publish is frozen before kick-off and graded that night against the closing line, market by market, and it all goes on the record page including the markets where the close beats us. A probability is not a tip. 18+, and please gamble responsibly.