Ultra-long Gamma-Ray Bursts are a class of high energy transients lasting several hours. Their exact nature is still elusive, and several models have been proposed to explain them. Because of the limited coverage of wide field gamma-ray detectors, the study of their prompt phase with sensitive narrow-field X-ray instruments could help in understanding the origin of ultra-long GRBs. However, the observers face a true problem in rapidly activating follow-up observations, due to the challenging identification of an ultra-long GRB before the end of the prompt phase. We present here a comparison of the prompt properties available after a few tens of minutes of a sample of ultra-long GRBs and normal long GRBs, looking for prior indicators of the long duration. We find that there is no such clear prior indicator of the duration of the burst. We also found that statistically, a burst lasting at least 10 and 20 minutes has respectively 28% and 50% probability to be an ultralong event. These findings point towards a common central engine for normal long and ultra-long GRBs, with the collapsar model privileged.