Why most prediction sites never say "under"
Overs sell. An over 2.5 goals tip promises action, it is easy to write up, and a site that only publishes overs never has to explain a 0-0 to anyone. That is why almost every tip page you will find lists the same thing: a column of fixtures with "Over 2.5" beside each one.
The trouble is that unders are half the market. Cagey fixtures, defensive form, low-scoring leagues and mid-table games with nothing at stake produce two goals or fewer about as often as open games produce three or more. A model that can only say "over" has thrown away every one of those fixtures before it starts.
The ScoreSage over/under board makes both calls from the same model. Under 2.5 selections sit on the same page as overs, can be filtered on their own, and carry their own settled record. This article explains how an under call is made, what the fields on the board mean, and how to read the record honestly.
What "under 2.5 goals" means
Under 2.5 goals wins if the match finishes with two goals or fewer in total: 0-0, 1-0, 0-1, 1-1, 2-0 or 0-2. Three goals or more loses it. Who scores does not matter, and neither does the result.
Because the line sits between two and three goals there is no push. That makes it a clean yes-or-no probability, which is exactly what a statistical model is good at estimating.
How the Total Goals model makes a call
The board is driven by the Total Goals engine, which runs on every fixture in the covered leagues before kick-off. Its output for a match is a single number, the expected total goals, and everything else follows from it.
1. Estimate the expected total. The model combines each side's scoring and conceding rates this season with the league's overall goal environment. A team that scores 1.1 a game against a defence conceding 0.9 a game in a league averaging 2.6 goals does not simply add up to 2.0; the rates are scaled against the league baseline so that a strong attack meeting a weak defence pushes the total up and a weak attack meeting a strong defence pulls it down. When a club has too few matches this season, the model falls back to its last five results and, failing that, to a league baseline, and the board labels which basis was used.
2. Turn the total into a probability. Goals in a football match are close to a Poisson process, so the expected total gives a probability for every scoreline. Summing the probabilities of zero, one and two goals gives P(under 2.5); one minus that is P(over 2.5). An expected total of 2.3 goals corresponds to roughly a 60% chance of under 2.5; an expected total of 3.0 corresponds to roughly 42%.
3. Anchor to the market where one exists. When a bookmaker has priced the over/under 2.5 line, the model de-vigs that price into a market probability and blends it with its own estimate. The market knows things a season-rate model does not, such as a confirmed lineup missing two forwards. Anchoring stops the model fighting the whole market on stale data.
4. Only call it when it is a call. If the resulting probability sits between 42% and 58%, the fixture is a coin flip and the board says so: it is listed with no verdict, for transparency, and never sold as a pick. Outside that zone the side with the higher probability becomes the call: OVER 2.5 or UNDER 2.5.
5. Grade the strength. The tier measures how far the probability is from 50%. Elite and strong calls are the ones that feed Best Bets and the settled ledger; moderate and low calls are shown so you can see the model's full view, but they are weaker by construction.
Reading an under pick on the board
Each row on the over/under board shows the same fields:
- The call, UNDER 2.5 or OVER 2.5, with the model's probability for it. A 64% under is the model saying the match finishes with two goals or fewer about two times in three.
- Expected total goals for the match. Unders with an expected total near 2.0 are stronger than unders at 2.4; the number tells you how far below the line the model sits.
- Tier: elite, strong, moderate or low.
- Basis: season, last five, or baseline. Treat baseline calls with more caution.
- Kick-off and league, and after the match a settled result: call won or call lost.
Use the filter to see only unders. The win rate shown at the top of the filtered board is the under 2.5 record for the dates you are looking at.
What the record says so far
Every call is settled automatically on the final score after full time, and the settled picks are counted on the verified results page. As of 8 September 2026 the under 2.5 market stands at 135 won and 68 lost, a 66.5% hit rate across 203 settled calls since the first under was published on 6 June 2026. Over 2.5 stands at 227 won and 105 lost, 68.4% across 332.
Two cautions. First, 203 calls is a real sample but not a large one; a hit rate can move several points on a bad fortnight. Second, a hit rate on its own says nothing about profit. A 66.5% strike rate breaks even at decimal odds of about 1.50. Under 2.5 is often priced anywhere from 1.60 to a little over 2.00 depending on the fixture, so the record is worth something only if you are backing calls at prices above the break-even point. The odds converter turns any price into the strike rate you need.
When unders are strongest
The league you are in matters more than any single team stat. The 2026/27 Season Hub ranks every club in 32 leagues on under 2.5 rate, with the league average alongside, so you can see where low-scoring football is normal rather than an exception. Serie A and Ligue 1 have historically cleared over 2.5 less than half the time; the Eredivisie and Bundesliga are the wrong place to look for unders.
Within a league, the model's strongest unders come from fixtures where both sides carry high clean-sheet rates and low expected totals, and where the market agrees. A lone under in a league that is scoring freely is a weaker proposition than the same probability in a league that is not.
Early in a season, be careful with anything built on three or four matches. The basis label exists for exactly this reason.
How to use it
- Compare the probability with the price. A 62% under at 1.55 is a worse bet than a 58% under at 1.85.
- Prefer elite and strong tiers. Moderate and low calls are shown for transparency, not as selections.
- Do not stack four unders from the same weekend into an accumulator and call it safe. Each leg is a probability; four of them multiply.
- Judge the board on its settled record over months, not on last night.
- Stake small and fixed. The record includes losing runs of a dozen calls, and it always will.
Where to find it
Today's calls are on the over/under board. The settled ledger is on verified results, and the league context for any fixture is in the Season Hub. For how the wider prediction models are built and tested, read how AI football predictions work; for how every pick is settled, read how the public record is settled.