Plenty of sites sell “AI picks” as magic. Ours is not magic — it is a transparent points model, and we would rather you knew exactly how it works. Here is the whole recipe.
The formula
Every player gets a score from 0 to 100, rebuilt every hour from official FPL data. Six ingredients, fixed weights:
- Form — 25%. Recent output, as published by FPL. Out of season we substitute points-per-game so the rankings stay meaningful.
- Attacking stats — 25%. Expected goals (40% of this slice), expected assists (30%) and the ICT index (30%). All from the official FPL API — no paid feeds.
- Fixtures — 20%. Our Fixture Score: the official difficulty of the next five matches, inverted to 0–100.
- Minutes security — 15%. How close the player is to 2,700 minutes (30 full matches). Bench players cannot fake this.
- Availability — 10%. Injury and suspension status straight from FPL, including the percentage chance of playing.
- Value — 5%. Points per million spent, so a £5m starter gets credit a £14m star does not need.
From rating to verdict
90+ is a Strong Buy, 80–89 Buy, 70–79 Hold, 60–69 Risky, below 60 Sell. A separate Watchlist tag is applied by hand when we see something the numbers have not caught up with yet. Yes — a human can overrule the machine, and every override is visible.
What the model cannot see
Press-conference whispers, tactical tweaks, a manager’s grudge. Percentile-based scoring also compresses differences at the very top. That is why every rating ships with a human-written “Why?” and why our editorial verdicts can differ from the raw number. Trust the model for the shortlist; trust your judgement for the final call.
Research and data analysis supported by AI (our own rating engine); reviewed and edited by the author. Ratings update hourly — check the live AI Player Rankings.