Graph Contrastive Learning (GCL) has been widely used in the field of rumor detection due to its ability to self-supervise the extraction of low dimensional representations of propagation structure. However, existing methods focus on augmenting propagation structures and applying GCL at different levels for events, ignoring the conflict between InfoNCE-based contrastive loss and the message-passing mechanism in Graph Neural Networks (GNNs). This conflict prevents negative samples from effectively pulling away from each other, resulting in poor performance in rumor detection. To address this problem, we propose a novel GCL method for rumor detection called RERG. RERG detects gradient confusion to capture the conflicts and then ignores negative samples that cause conflicts, allowing GNNs to learn from the ignored negative samples adaptively. In addition, we propose a propagation structure augmentation method that considers edge importance to preserve the critical information of the propagation structure, thereby obtaining high-quality positive samples. Experimental results on public datasets prove that RERG achieves better results than other state-of-the-art models.