A key component of graph and relational learning methods is the computation of vector representations of the input graphs or relations. The starting point of this tutorial is that we model this computation as queries, mapping relational objects into the realm of real vector spaces. We then revisit recent works in the machine learning community on the expressive power of graph learning methods from this unifying query language perspective. Here, we consider the expressive power related to the discrimination of inputs and to the approximation power of functions. Finally, we argue that the bridge between graph learning and query languages opens many interesting avenues for further research.