The current surge of interest in graph-based data models mirrors the usage of increasingly complex reachability queries, as witnessed by recent analytical studies on real-world graph query logs. Despite the maturity of graph DBMS capabilities, complex label-constrained reachability queries, along with their corresponding aggregate versions, remain difficult to evaluate. In this paper, we focus on the approximate evaluation of counting label-constrained reachability queries. We offer a human-explainable solution to graph Approximate Query Processing (AQP). This consists of a summarization algorithm (GRASP), as well as of a custom visualization plug-in, which allows users to explore the obtained summaries. We prove that the problem of node group minimization, associated to the creation of GRASP summaries, is NP-complete. Nonetheless, our GRASP summaries are reasonably small in practice, even for large graph instances, and guarantee approximate graph query answering, paired with controllable error estimates. We experimentally gauge the scalability and efficiency of our GRASP algorithm, and verify the accuracy and error estimation of the graph AQP module. To the best of our knowledge, ours is the first system capable of handling visualization-driven approximate graph analytics for complex label-constrained reachability queries.