Frequently asked questions
What the numbers mean, where they come from, and how much to trust them. If your question isn't here, ask — it probably belongs on this page.
What is FPL Buddy?
A free set of Fantasy Premier League tools. It rates every player with a trained points model, picks two AI squads each gameweek, finds pairs of clubs whose fixtures rotate well, and analyses your own squad from your FPL ID. It is an unofficial fan project with no connection to the Premier League.
How do I find my FPL ID?
Log in at fantasy.premierleague.com, open the Points tab, and look at the address bar. The URL reads /entry/1234567/event/5 and your ID is the number after /entry/. It is public, it is not a password, and it is safe to share.
Does FPL Buddy need my Fantasy Premier League password?
No, and it never will. Everything here uses the public read-only FPL API, which needs nothing but your numeric team ID. That also means the site cannot make your transfers for you — it can only recommend them, and you apply them yourself in the official app.
How are the predicted points calculated?
Two gradient-boosted models per position, because goalkeepers, defenders, midfielders and forwards earn points in different ways — and because "how many will he score" is really two questions. Just over half of all gameweek rows are players who didn't play, so one model across that mixture mostly learns who plays and flattens everything else. Instead: a classifier for whether he starts, and a Poisson model for what he scores when he does, fitted only on appearances. Inputs are FPL's own per-gameweek columns over a three- and a six-gameweek window plus a season-to-date rate, with home or away and opponent strength. All of them are shifted back a gameweek so a player's own result can never leak into the features predicting it. Full detail on the about page.
How accurate are the predictions?
Useful for comparing players against each other; poor as a forecast of any single score. Football over ninety minutes is not very predictable, and a model working from a handful of inputs will be wrong about individual players most weeks. The honest use is "who is the better pick", not "how many will he get". You do not have to take that on trust: the accuracy page publishes every frozen prediction against what the player actually scored, including how the model compares with not modelling at all.
My team rating went up but I scored fewer points. Why?
Because the rating isn't a points forecast. A player's rating out of 100 is his projected points ranked against the other players in his position, so the best goalkeeper and the best forward both read 100 while being worth very different scores. Your team rating averages that across the starting eleven, which measures squad quality and has no points meaning. It also describes the players as rated today rather than for one gameweek, and it ignores your captain. The predicted points figure next to it is the number that answers "how many this week". More on the about page.
What is the difference between AI Best XI and AI Manager?
AI Best XI is rebuilt from scratch every gameweek with no memory: the highest-scoring legal £100m squad for the week ahead, spending almost nothing on the bench because it has no transfers to plan for. AI Manager plays the actual game — one squad across the season, a bank balance, free transfers, four-point hits when a move earns one, and a single set of chips — so it judges players over several gameweeks rather than one.
Why do the two AI squads disagree?
Because they are answering different questions. Buying the best captain for one Saturday and selling him a fortnight later costs a hit and a free transfer, which Best XI never pays and AI Manager always does. When they agree on a player, that is a stronger signal than either on its own.
How does the fixture rotator work?
It looks for pairs of clubs whose fixtures alternate, so that in most gameweeks at least one of the two has a favourable game. Own a defender from each and you can start whichever has the better fixture that week. Pairs are ranked over the next eight gameweeks, separately for defence and attack.
What does "underperforming" mean on the players page?
A player whose returns are behind their underlying numbers: attackers scoring fewer goals than their expected goals suggest, or goalkeepers and defenders conceding more than their expected goals conceded. It is a list of players who may be due a correction, not a list of players who definitely are.
Is a particular player injured, and when is he back?
The injury news page lists every player Fantasy Premier League currently flags — injured, suspended or doubtful — with the game's own chance-of-playing percentage where it gives one. Return dates are not guessed at: if FPL hasn't published a number, neither does this site. Each player's own page carries the same status alongside his form and projections.
What does "overperforming" mean on the players page?
The same comparison the other way round: attackers scoring more than their expected goals, or defences conceding fewer than their expected goals conceded. It is not a sell list. A player can stay ahead of his underlying numbers for a long time — good finishers do it for whole seasons — so read it as a run that may not hold rather than one that is about to end.
What is a differential, and how do I find one?
A player owned by a small share of managers, so that when they return points you gain ground on almost everyone. Sort the player ratings by predicted points and look for low ownership — a high projection at low ownership is the combination worth having. Once you have two or three names, the comparison tool puts them side by side on projected points, value per million and fixtures.
When do FPL player prices change, and can you predict them?
Prices move overnight, at around 01:30 UK time. A player rises when enough managers transfer him in and falls when enough transfer him out, and the number needed scales with how many people already own him — but Fantasy Premier League has never published the actual rule, so every prediction of it, here and everywhere else, is reverse-engineering. The price change page shows who is closest and, unusually, shows the evidence it is working from: the threshold it uses is measured from real changes it has watched happen, and it says how many that is. Worth a look before a transfer, not worth taking a hit for — you only get half of a price rise back when you sell.
How often is the data updated?
The full pipeline runs nightly at 03:00 UK time: prices, ownership, form, injuries, fixtures and every projection. A lighter check runs hourly to catch deadlines and freeze the AI squads' predictions before each one. Live scores update during matches.
Is FPL Buddy free?
Yes, entirely, with no account and no sign-up. There is a Ko-fi link in the footer if you want to help with the server bill, but nothing on the site is behind it.
Still stuck?
Ask here. Nothing is stored — not an email, not an address, not a name — so this cannot be replied to directly. What happens instead is that questions worth answering end up on this page, where the answer reaches you and everyone who asks it next.
If you want an actual reply — a bug, a player whose numbers look wrong, or a disagreement about the model — the contact page has the address. The about page goes into how the model is trained and what it deliberately leaves out.