Do earthquakes start with a precursory phase of slow slip? To investigate this question, we compiled all the GPS data recorded before large (Magnitude ≥ 7.0) earthquakes, projected them in the direction of expected displacements, and "stacked" (i.e. summed) everything. The result is a time series showing that on average the consistency with precursory slow slip increases in the 2 hours preceding the ruptures (Bletery and Nocquet, Science, 2023). This cannot be used to predict earthquakes but suggests that earthquakes do start with a precursory phase of slow slip on the fault. This signal is rather subtle though and this result will need to be confirmed by future studies.