Conventional co-salient object detection (CoSOD) has a strong assumption that a common salient object exists in every image of the same group. However, the biased assumption contradicts real scenarios where co-salient objects could be partially or completely absent in a group of images. We propose a random sampling based Generalised CoSOD Training (GCT) strategy to distill the awareness of inter-image absence of co-salient object(s) into CoSOD models. In addition, the random sampling process inherent in GCT enables the generation of a high-quality uncertainty map, with which we can further remediate less confident model predictions that are prone to localising non-common salient objects. To evaluate the generalisation ability of CoSOD models, we propose two new testing datasets, namely CoCA-Common and CoCA-Zero, where a common salient object is partially present in the former and completely absent in the latter. Extensive experiments demonstrate that our proposed method significantly improves the generalisation ability of CoSOD models on the two new datasets, while not negatively impacting its performance under the conventional CoSOD setting. Codes are available at https://github.com/Carlisle-Liu/GCoSOD.