Abstract | In performance‐based earthquake wngineering, record selection comes into play at the interface of seismic hazard and structural analysis aiming to repair any loss of essential seismological dependencies caused by the choice of an insufficient intensity measure to be used for structural response prediction. Site‐specific selection is best exemplified by the prominent conditional spectrum (CS) approach that attempts to ensure a hazard‐consistent response prediction by involving site hazard disaggregation. Specifically, CS utilizes a target spectrum (with mean and dispersion) that, in its latest formulation, accounts for all the scenarios (in terms of magnitude, M, and closest to rupture distance, R) contributing to the hazard of the site at a given intensity level. The ground motion records, however, are selected to match this target spectrum–based solely on their spectral shape but with no explicit consideration to their underlying M‐R characteristics. The main focus of this study is to explore whether the reintroduction of M‐R criteria in the selection process preserves hidden dependencies that may otherwise be lost through a spectral‐shape‐only proxy. The proposed record selection method, termed CS‐MR, offers a simple approach to maintain a higher order of hazard consistency able to indirectly account for metrics that depend on M‐R (e.g., duration, Arias intensity) but are not captured in the response spectra. Herein the CS‐MR response prediction is favorably compared to CS and to the generalized conditional intensity measure methods that select records according to, respectively, spectral shape only and, for the case at hand, to spectral shape plus duration.