Observations of streaming potential for unsaturated conditions do not always show the same trend depending on the hydrodynamic conditions and because of a lake of coherency between the data processing procedures. We combine the data from three studies published in the literature, acquired during non-steady state drainage experiments, and apply the same processing steps. We model the hydrodynamic behaviour of these experiments to confirm that they experienced different flow dynamics. We argue that the raw SP data should not be corrected unless a clear drift of the electrodes stability is observed. The combined hydrodynamic behaviour and the streaming potential response show that (a) the observations of one of the experiment (exp #1) are associated to a limited range of water saturation (0.85-1). The corresponding signals could 16 therefore be fairly modelled assuming no saturation dependence of the SPC whatsoever; (b) the observations of exp #3 led to a SPC that can be larger than its value at saturation; (c) the observations of the exp #2 show a non-monotonous behaviour of the SPC as saturation decreases. The underlying physics of a non-monotonous SPC is related to water/air interfaces as shown by the results of the lattice Boltzmann numerical simulations. The main contribution to the SPC behaviour comes from the charged water/air interfaces and depends on the dynamic state of moving or entrapped bubbles. We finally describe the consequences of such a behaviour on the seismoelectric conversions for unsaturated conditions.