How AIGround predicts
AIGround turns a game into a short, evidence-based briefing before it starts. This page explains exactly what happens under the hood, what the numbers mean, and — just as importantly — what our predictions are not.
1. When a prediction is generated
Each prediction is generated once per game, about three hours before first pitch or kickoff. Generating close to game time means the model can use the latest probable starters, confirmed lineups, injuries, and form rather than a stale forecast from days earlier. The result is cached and served to every visitor, so everyone reads the same analysis — we don’t re-generate it per view.
2. Grounding: the model searches before it reasons
The model does not predict from memory. For every game it first searches current, game-specific sources — official league records, sports statistics sites, and same-day preview coverage — and cites concrete figures from them. For baseball that means both starting pitchers’ ERA and WHIP, recent outings, team last-10 records, the season series, bullpen fatigue, and the actual ballpark and its park factors. For World Cup football it means likely formations and starting XIs, last-5 form, FIFA ranking, key players, head-to-head history, group standings, and the host city’s conditions.
3. The numbers-first principle
Every key factor must carry at least one concrete number — an ERA, a last-10 record, a head-to-head tally, a FIFA ranking. Vague phrases like "historically strong" or "good momentum" are not allowed; they have to be expressed as a figure or a fact. If a number can’t be verified by search, the model is instructed not to invent it — it says the value is unconfirmed rather than guessing.
4. What you get on each game page
- Predicted winner and a win probability calibrated to the strength gap (a coin-flip sits in the low 50s; a clear favorite 65%+).
- A projected score estimated from scoring and run-prevention indicators, kept consistent with the predicted winner.
- Key factors — the numbered evidence behind the call.
- A briefing: a scouting-report style write-up that ties the factors together.
After the game goes final, each page shows whether the pick was correct, missed, or a draw — we don’t hide the misses.
5. Live data
Scores, line scores, box scores, and match timelines are updated live from sports data providers while a game is in progress. For MLB, your browser also reads MLB’s official StatsAPI directly to show pitch-by-pitch detail and a live win probability. This live data is factual game state, separate from the pre-game AI briefing.
6. Limits — please read
Sports are uncertain, and AI predictions are estimates, not outcomes. Even a well-grounded 65% pick is expected to be wrong about a third of the time — that is what the probability means. Everything here is for information and entertainment. AIGround does not encourage betting or gambling, and you should never treat a prediction as a wagering signal.