Scaling relations play an important role in fault‐based seismic hazard estimates. Scaling relations impact estimates of the sizes and rate for a given fault area. Here, we examine which relations are the most useful for these estimates and issues that arise. These include the depth of large event ruptures, transient deepening of seismicity following large events, difficulties in using classical continuum exponent fits, and the importance of large event asymptotics. A new analysis of land‐based data calls into question nonspecific exponent fitting, which is a standard practice. We show a dependence on the lower and upper cut‐off magnitudes in the data in the best‐fitting slope parameter relating magnitude to log area with this approach. We show as well a dependence on assumed data uncertainties. These sensitivities make using this quite standard approach very problematic. Based on this evidence and other factors, we propose recommendations for the minimal branch sets, which preserve epistemic uncertainty for use in fault‐based seismic hazard estimates.