Graph Auto-Encoders (GAEs) have demonstrated remarkable effectiveness in detecting anomalies in graphs. However, their "black-box" nature makes it difficult to understand why they classify a node as anomalous. Moreover, despite the development of XAI, where many methods have been proposed to provide explanations for various deep learning models, there is a notable lack of an evaluation framework dedicated to anomaly detection in graphs. Our contribution addresses this gap by adapting existing evaluation frameworks to the specific challenges of anomaly detection using GAEs. Additionally, it introduces a simple yet effective explanation technique based on GAE reconstruction errors. Using this new framework, we evaluate the effectiveness of different explainers and experimentally show that the method we propose, based on reconstruction errors, outperforms other explainers for GAEs.