AI predictions
August 21, 2026
Predictions are at the heart of SignalTurf.
For each race, the app displays a selection of horses based on the analysis carried out by its artificial intelligence models. These predictions are calculated automatically before each race starts, based on the information available at that time.
Their goal is to highlight the horses that stand out most from the models’ analysis and to give you a starting point for your own thinking.
Predictions generated automatically
Unlike predictions written by an expert, the selections offered by SignalTurf are produced entirely automatically.
The artificial intelligence models analyze each race using an identical method, with no human involvement in choosing the horses.
This approach ensures that every race is evaluated using the same criteria, regardless of the racecourse, discipline, or horses entered.
A new analysis for every race
Predictions are never reused from one race to another.
Every race has its own characteristics: competitors, distance, discipline, and racing conditions all change constantly.
For this reason, the models run a new calculation before every start to produce an analysis specific to that race.
Each selection therefore only applies to the race shown in the app.
What the selected horses represent
The highlighted horses are the ones that stand out most from the analysis carried out by the models.
In other words, they correspond to the competitors showing the most favorable characteristics according to the evaluation method used by SignalTurf.
However, this isn’t a final ranking or a certainty about the race’s outcome.
The other horses remain capable of a stronger performance, as in any sporting competition.
A tool to support your analysis
Predictions should be seen as a decision-support tool.
They help guide your reading of the race by highlighting the horses that draw the most attention from the artificial intelligence models.
They don’t replace your own analysis and don’t constitute personalized advice or a guarantee of results.
Every user remains free to interpret the information offered according to their own approach.
Why predictions can change
A race can change up until a few minutes before it starts.
For example, a horse may be declared a non-runner, or certain official information may be updated.
When these changes are likely to affect the analysis, SignalTurf may automatically recalculate the predictions to incorporate the most recent information.
It’s therefore possible for the selected horses to change between the analysis’s first publication and the race’s actual start.
Not all predictions are equal
Some races produce a very consistent analysis, while others are much harder to interpret.
That’s why SignalTurf complements its predictions with a confidence level.
This indicator adds extra context to the displayed selection. Two races may offer a similar number of selected horses while showing different confidence levels.
It’s recommended to interpret predictions with this indicator in mind, rather than looking only at the list of selected horses.
Looking at performance over time
A prediction should never be judged based on a single race.
Like any analysis method, the models have both successes and failures.
That’s why SignalTurf provides historical statistics that let you evaluate model performance across a large number of races, rather than a few isolated examples.
This approach offers a more representative view of how the analyses actually behave.
The limitations of predictions
No artificial intelligence model can predict the outcome of a horse race with certainty.
Horses’ performance remains influenced by many factors, and a degree of uncertainty always remains.
The predictions offered by SignalTurf represent the most relevant analysis produced by the models at the time of calculation, but they never constitute a promise of winnings or a guarantee of victory.
They should be used as one piece of information among others.
Go further
You now know what the predictions shown in SignalTurf represent and how to interpret them.
The AI models chapter presents the approach used to produce these analyses in more detail, while Confidence level explains how to complement your reading of predictions with an additional indicator.
If you’re looking for the definition of a specific term, you can also check the Glossary or the FAQ.