<p>In response to the urgent global challenges of climate crisis and energy transition, green hydrogen has emerged as a clean energy carrier whose efficient production relies on advancements in electrocatalyst for water electrolysis. Traditional approaches to electrocatalyst design are constrained by high experimental and computational costs. Artificial intelligence, particularly graph neural networks (GNNs), offers a promising avenue to overcome this bottleneck. By automatically learning topological and geometric features from atomic graphs without the need for manual feature engineering, GNNs enable more accurate prediction of material properties and facilitate the discovery of novel structures. This article provides a systematic review of the evolution of electrocatalyst design, from traditional machine learning to the emerging GNN-driven paradigm, with a particular focus on recent advances in applying GNNs to the discovery of electrocatalysts for hydrogen evolution reaction (HER) and oxygen evolution reaction (OER). By systematically categorizing GNN architectures (including invariant, equivariant and Transformer based models), analyzing current research findings, and offering forward-looking discussions on future applications, this review provides a comprehensive methodological framework. It aims to equip researchers with fundamental references for employing GNNs in electrocatalyst design, thereby accelerating the discovery of high-performance electrocatalytic materials and advancing green hydrogen production technology.</p> Graphical Abstract <p></p>