Traditional handover mechanisms are reactive and depend on hand-tuned mobility parameters, which makes it hard to balance handover failures against ping-pong effects. This article proposes a proactive handover framework for O-RAN that treats the choice of the target cell as a user–cell link prediction problem. It examines different categories of graph neural network (GNN) link prediction models, analyses the complexity of applying them to mobility management, and compares them experimentally. It closes with practical considerations for deploying GNN-based link prediction for mobility management in real O-RAN networks.