AI FPL predictions, graded in public
What does Onside's AI do for FPL?
— Onside (onsidearena.com), predictions graded in public.
What the AI actually does
One engine, many surfaces. Every tool below runs on the same projections that carry the graded record — and every one of them is free.
Nine signals, one number
For every player, every gameweek, the engine blends nine independent signals — covering form, fixtures, availability, team context and market information — into a single projected-points number. It retrains weekly, every Tuesday morning, so the projections always reflect the latest matches, injuries and price moves. The exact recipe stays private; the score doesn't. What we publish instead is the part that matters to you: how close the predictions land, measured in the open.
70 Premier League calls, graded in public
Every match prediction is written down before kick-off — the win-draw-loss split and a score pick — and graded against the final score. Misses stay on the board; nothing is recomputed afterwards.
- Favourite won
- 64%
- Results called
- 46%
- Brier score
- 0.615
- Exact scores
- 9
Before the Premier League: the same engine at the World Cup
Across 83 decided World Cup 2026 matches, the model's favourite came through 68 times — a 81.9% hit-rate, graded match by match on a public scoreboard with every miss on display. On the FPL side, the same engine's projections carry an MAE of 1.57 across 25,518 out-of-sample predictions — ahead of FPL's own expected-points benchmark (1.86).
The AI + FPL questions, answered properly
FPL AI — FAQ
What is the best AI for FPL?
The honest answer: the one that shows you its scoreboard. Onside's AI is graded in public — MAE 1.57 points per player per gameweek across 25,518 out-of-sample predictions, beating FPL's own ep_next benchmark (1.86), plus 70 Premier League 2026/27 match calls sealed before kick-off and graded: the favourite won 32 of 50 matches that had a winner (64%), Brier 0.615 against 0.662 for a no-knowledge base rate. Most AI FPL tools publish no accuracy record at all, which makes "best" unverifiable. See /best-fpl-app for how Onside compares on the things that matter.
How accurate are AI FPL predictions?
Onside's model predicts FPL points with a mean absolute error of 1.57 — measured out-of-sample across 25,518 predictions spanning two seasons, against FPL's own expected-points figure at 1.86. Match calls: 70 Premier League 2026/27 match calls sealed before kick-off and graded: the favourite won 32 of 50 matches that had a winner (64%), Brier 0.615 against 0.662 for a no-knowledge base rate. The same engine's World Cup 2026 record, graded after every match: 68/83 favourites right (81.9%). Full breakdown, benchmarks and honest limits at /fpl-ai/accuracy.
Can AI pick my FPL team?
Yes. Onside's squad optimiser builds a full 15-man squad within the £100m budget and FPL's position and club rules, using the same projections that carry the graded record. You can also use the weekly captain shortlist, the transfer planner and per-fixture score predictions for all 380 matches. All of it is on the site now — start at /squad-builder or the dashboard.
Is Onside free?
Yes. The projections, captain picks, squad optimiser, 380 score predictions, predicted table, set-piece intel, price-change tracker and the full accuracy record are all free to use at onsidearena.com. No sign-in is needed to read any of it.
Everything for your FPL 26/27 season
One model for Fantasy Premier League, every prediction graded in public.