NMR spectroscopy is a excellent tool for monitoring in-situ chemical reactions. In particular, DOSY measurement is well suited to characterize transient species by the determination of their sizes. However, here we bring to light a difficulty in the DOSY experiments performed in out-of-equilibrium systems. On such a system, the evolution of the concentration of species interferes with the measurement process, and creates a bias on the diffusion coefficient determination that may lead to erroneous interpretations.We show that a random permutation of the series of gradient strengths used during the DOSY experiment allows to average out this bias. This approach, that we name p-DOSY does not require changes in the the pulse sequences nor in the processing software, and restores completely the full accuracy of the measure. This technique is demonstrated on the monitoring of the anomerization reaction of α-to β-glucose.