ChatGPT for FPL vs a purpose-built model
This is not a hit-piece — general AI assistants are genuinely useful for FPL, and this page says exactly where. It also says where they structurally can't compete with a model built for the job, and why the smart answer is to use both.
ChatGPT or a dedicated FPL model?
What general AI assistants genuinely do well
What a chat model can't do by itself
None of this is a criticism of the technology — these are different tools. A conversational model is built to reason in language; an FPL model is built to turn live football data into graded probabilities. The gaps close only when the assistant is connected to a live, purpose-built source.
What Onside brings to the table
Onside is the other half of the stack: live FPL data refreshed continuously, per-player projections with a published MAE of 0.86 (against FPL's own 0.896) across 51,518 out-of-sample predictions, calibrated probabilities you can check on the calibration page, a squad optimiser that solves the £100m problem properly, and a public graded record — 81.9% across 83 decided World Cup 2026 matches, misses included.
The twist: AI assistants already use Onside
The best version of "ChatGPT for FPL" is an assistant grounded in real, graded data — and Onside is built to be that ground truth. The site publishes an llms.txt guide for AI crawlers, open JSON endpoints anyone can call without a key, and an MCP server (onside-football-mcp on npm) that tool-using assistants can query directly. Ask a connected assistant for this week's captain or a match prediction, and it can answer with Onside's live numbers — the conversational layer you like, on top of the graded model you can verify.
Use both — for what each is for
Talk your decisions through with a general assistant. Take your numbers from a model with a public record. If the assistant can call live data, point it at Onside; if it can't, sanity-check its tips against the 380 score predictions and the captain shortlist before you hit confirm. Everything on the Onside side is free.
ChatGPT + FPL — FAQ
Can ChatGPT give good FPL tips?
Partly. ChatGPT and similar assistants are genuinely good at explaining FPL rules, chip strategy and talking through decisions — as a sounding board they are excellent. What they lack by default is live data (prices, fixtures, injuries, ownership), calibrated probabilities, and any graded record of past tips. For "who will score the most points this week", they are guessing more than modelling — unless they are grounded in live data from a purpose-built source.
Should I use ChatGPT or an FPL prediction model?
Both, for different jobs. Use a general assistant to think out loud — rules, scenarios, reasoning. Use a purpose-built, graded model for the numbers: projections, captain probabilities, squad optimisation. Onside's engine carries a published MAE of 0.86 across 51,518 out-of-sample predictions and a 81.9% graded World Cup 2026 hit-rate — a record a conversational model simply doesn't have.
Can ChatGPT use Onside data?
Yes — this is the good news for AI-assistant users. Onside publishes its FPL and match data for AI consumption: a llms.txt guide, open JSON endpoints, and an MCP server (onside-football-mcp on npm) that ChatGPT- and Claude-style clients can call directly. Ask an AI assistant with web or tool access about FPL captain picks or match predictions and it can ground its answer in Onside's live, graded numbers rather than guessing.
Why does a graded record matter for FPL advice?
Because anyone can sound confident. FPL advice is only worth following if the source is right more often than the alternatives, and the only way to know that is a public, out-of-sample record with the misses included. Onside grades itself in the open — every World Cup 2026 call is on the scoreboard, and FPL projection accuracy is published against FPL's own benchmark. Advice without a record is entertainment; advice with one is information.
Everything for your FPL 26/27 season
The same model that graded every World Cup call in public, pointed at Fantasy Premier League.