Aims. We investigated the impact of photometric redshift errors in the ongoing Javalambre Physics of the Accelerating Universe Astrophysical Survey (J-PAS) on void identification and void properties using a watershed-based method. Our aim is to assess the recovery of individual voids and the overall void environment. Methods. We created galaxy mock catalogues for a redshift of z = 0.1, using the IllustrisTNG300-1 simulation and defining two datasets: an ideal sample (mr < 21 mag) and a perturbed sample with the Z-coordinate errors mimicking J-PAS's line-of-sight errors, derived from the precursor miniJPAS survey data. We identified voids using the watershed algorithm ZOBOV. Results. We found 1065 voids in the ideal sample and 2558 voids in the perturbed sample. The perturbed sample voids have, on average, smaller sizes and denser interiors. We filtered out voids based on density and radius to eliminate overdense and small spurious instances. The stacked density profile of filtered voids in the perturbed sample remains close to the average density, even at the boundary peak, indicating a strong blurring of structures by the redshift errors. The number of the ideal sample voids for which at least 50% of the volume is recovered by a void in the perturbed sample is 53 (29 for the filtered sample). The volume occupied by these voids is less than 10% of the simulation volume. Merging voids in the perturbed sample marginally improves the recovery. The overall volumes defined as voids in the two samples have an overlap of 80%, making up 61% of the simulation box volume. Conclusions. While some statistical properties of voids might be recovered sufficiently well, the watershed algorithms may not be optimal for recovering the large-scale structure voids if they are applied directly to photometric redshift survey data.