This page explains, in plain language, how a Gopredictfy prediction is produced, how we check it, and where it can go wrong. If you only read one page on the site before deciding whether to trust it, make it this one.
For every match we estimate the probability of each result and of common goal markets using a statistical model built on recent results. We compare that estimate with independent sources and with bookmaker prices, adjust our stated confidence using how accurate we have really been in the past, and publish the pick before kickoff. After the match we mark it right or wrong and add it to a public record.
The model works from information that is available before a match starts: each team’s last several results, the goals it scores and concedes per game, its home and away record, a rating that measures overall strength and moves after every result, and the head-to-head record between the two clubs.
It does not use inside information, and it does not include team news, injuries or line-ups unless they have been reflected in the data feeds we receive. That is a genuine limit, and it is why every preview we publish carries a reminder to check team news before kickoff.
At the centre is a goal-based model of the kind widely used in football analytics. It treats the number of goals each side scores as a random quantity whose average depends on the attacking strength of one team and the defensive weakness of the other, with an adjustment for home advantage. From those averages it can work out the chance of every scoreline, and from the scorelines it can add up the probability of a home win, a draw, an away win, Over or Under a goals line, and both teams scoring.
Low-scoring games are more likely to end level than a simple model suggests, so the model includes a correction for that. The strength of the correction is not fixed by hand: it is re-estimated from real results every week.
No single model is right all the time. Our public pick is a consensus that draws on our own model and on other independent prediction sources, and each source is weighted by how accurate it has actually been over a meaningful number of past predictions, not by reputation. If a source starts doing worse, its weight falls automatically.
We also look at bookmaker prices. They are not a crystal ball, but they reflect a great deal of information. When our estimate and the market disagree strongly, the disagreement is often noise rather than a real edge — a pattern we have measured on our own results — so we treat very large gaps with caution instead of chasing them.
Each prediction has a confidence figure. The aim is simple: among all the picks we label 70%, about 70% should turn out right; among those we label 85%, about 85%. Raw model output is not automatically like that, so every week we compare stated confidence with real outcomes, market by market, and correct the mapping. A market where we have too little data yet uses a cautious fallback until there is enough to learn from.
This is what we mean when we say the model learns from its results. It does not rewrite its own reasoning, but the way it converts an estimate into a confidence figure, and its baseline rates for each market, are updated from evidence every week.
We cover the match result (home win, draw, away win), Double Chance (two of the three results at once), total goals lines such as Over 1.5, Over 2.5 and Under 3.5, and both teams to score. Different markets have very different natural hit rates. Double Chance is right more often than a straight win pick simply because it covers two outcomes, so a fair comparison is always against that market’s baseline, not against a single number. Our accuracy page shows each market separately for exactly this reason.
We have also stopped publishing a market when the evidence showed it was not beating chance. Removing something that does not work is part of the method.
Every prediction is stored with its time of publication and marked correct or incorrect after the match. The accuracy page shows the totals, the split by market, and the daily record, including the bad days. We do not delete losing predictions or add winners after the fact.
Small samples mislead. A market with a handful of picks can show a perfect record or a terrible one by luck alone, so we flag small samples instead of presenting them as proof.
Football is a low-scoring game with a great deal of chance. Late injuries, red cards, weather, motivation and simple bad luck are not fully captured by any model. Predictions are estimates of what is more likely, never certainties, and a run of losses does not by itself mean the model is broken any more than a run of wins means it is perfect.
Predictions are provided for information and entertainment only. We are not a bookmaker and do not accept bets. If you bet, please do so responsibly, only with money you can afford to lose, and only if you are 18 or over.