A central property of a classical geometry is that the geodesic distance between two events is additive. When considering quantum fluctuations in the metric or a quantum or statistical superposition of different spacetimes, additivity is generically lost at the level of expectation values. In the presence of a superposition of metrics, distances can be made diffeomorphism invariant by considering the frame of a family of free-falling observers or a pressureless fluid, provided we work at sufficiently low energies. We propose to use the average squared distance between two events 〈d^2(x,y)〉 as a proxy for understanding the effective quantum (or statistical) geometry and the emergent causal relations among such observers. At each point, the average squared distance 〈d^2(x,y)〉defines an average metric tensor. However, due to non-additivity, 〈d^2(x,y)〉 is not the (squared) geodesic distance associated with it. We show that departures from additivity can be conveniently captured by a bi-local quantity C(x, y). Violations of additivity build up with the mutual separation between x and y and can correspond to C< 0 (subadditive) or C > 0 (superadditive). We show that average Euclidean distances are always subadditive: they satisfy the triangle inequality but generally fail to saturate it. In Lorentzian signature there is no definite result about the sign of C, most physical examples give C < 0 but there exist counterexamples. The causality induced by subadditive Lorentzian distances is unorthodox but not pathological. Superadditivity violates the transitivity of causal relations. On these bases, we argue that subadditive distances are the expected outcome of dynamical evolution, if relatively generic physical initial conditions are considered.