Majority of the existing graph neural networks (GNN) learn node embeddings\nthat encode their local neighborhoods but not their positions. Consequently,\ntwo nodes that are vastly distant but located in similar local neighborhoods\nmap to similar embeddings in those networks. This limitation prevents accurate\nperformance in predictive tasks that rely on position information. In this\npaper, we develop GraphReach, a position-aware inductive GNN that captures the\nglobal positions of nodes through reachability estimations with respect to a\nset of anchor nodes. The anchors are strategically selected so that\nreachability estimations across all the nodes are maximized. We show that this\ncombinatorial anchor selection problem is NP-hard and, consequently, develop a\ngreedy (1-1/e) approximation heuristic. Empirical evaluation against\nstate-of-the-art GNN architectures reveal that GraphReach provides up to 40%\nrelative improvement in accuracy. In addition, it is more robust to adversarial\nattacks.\n