Probabilistic seismic hazard analysis (PSHA) is the state‐of‐the‐art method to estimate ground motions exceeded by large, infrequent, and potentially damaging earthquakes; however, a fundamental problem is the lack of an accepted method for both quantitatively validating and refining the hazard estimates using empirical geological data. In this study, to reduce uncertainties in such hazard estimates, we present a new method that uses empirical data from precariously balanced rocks (PBRs) in coastal Central California. We calculate the probability of toppling of each PBR at defined ground‐motion levels and determine the age at which the PBRs obtained their current fragile geometries using a novel implementation of cosmogenic 10Be exposure dating. By eliminating the PSHA estimates inconsistent with at least a 5% probability of PBR survival, the mean ground‐motion estimate corresponding to the hazard level of 10−4 yr−1 (10,000 yr mean return period) is significantly reduced by 27%, and the range of estimated 5th–95th fractile ground motions is reduced by 49%. Such significant reductions in uncertainties make it possible to more reliably assess the safety and security of critical infrastructure in earthquake‐prone regions worldwide.