AI football predictions
FootIQ combines statistical models with an AI review layer. The mathematics produces the probabilities; the AI model, currently from OpenAI, checks each candidate against the full match context and can reject it or make it more cautious.
Why combine statistics and AI
Statistical models are consistent and measurable, but they only see what is in the numbers. An AI model can read the wider context of a match (absences, lineups, a team's schedule, unusual form) and flag cases where the numbers are likely to mislead.
What the AI model receives
For each match: the fixture and competition, every model's expected goals and probabilities, team strengths, recent form, home and away records, head-to-head results, xG, injuries and suspensions, lineups when announced, standings, bookmaker prices and data-quality measures. It receives no personal data about FootIQ users.
What the AI is allowed to do
- Accept a candidate pick as it is.
- Reject it, with a stated reason.
- Lower its probability, which may drop it below the market minimum.
- Never raise a probability. The final numbers always come from the calibrated statistical models, so the AI cannot make a pick look stronger than the data supports.
If no AI model is configured, or the model fails to respond, deterministic rules decide instead, and the pick is still subject to all statistical gates.
Live AI predictions
During matches, Premium members see live predictions on the live match cards. These are decided by the AI model from live data: the score, time, and live statistics such as shots and possession. It starts analysing from kick-off and publishes from the 10th minute. Each live pick states its scope (first half, second half, rest of the match or full match). It can be replaced after significant events such as a goal or a red card.
Limitations
- AI models can be wrong, and they can misread context; that is why they may only make picks more cautious.
- Live predictions depend on the timeliness of live data feeds.
- Neither the models nor the AI can know about events that are not in the data (for example a late illness).
Read the complete methodology for how the statistical part works.