Abstract Fossils are crucial for accurately dating phylogenetic trees because their ages provide vital constraints on the timing of macroevolutionary events, and their morphological characters offer key information on evolutionary rates and phylogenetic positions. The fossilized birth–death (FBD) process is a diversification model that incorporates both extant and extinct species, serving as a tree prior that seamlessly integrates fossils into phylogenetic inference. Although the FBD model can account for mass extinctions, which caused rapid, widespread organismal loss, few studies have utilized FBD models incorporating these events in phylogenetic inference. This is likely because the detectability of mass extinctions and their impact on phylogenetic inference remain unclear. Through simulations, we assessed the influence of mass extinctions on divergence time and topology inference and evaluated the detectability of mass extinction signals in total-evidence dating. We examined three FBD tree priors: without mass extinction, with known mass extinction time and survival probability, and with known mass extinction time but unknown survival probability. Our results show that the FBD model with known mass extinction time and unknown survival probability was able to reliably detect mass extinctions when they occurred, and correctly refrained from detecting mass extinctions when they were absent. Moreover, different FBD models generate similar divergence time and tree topology errors. Even when the FBD model used for tree inference did not explicitly account for mass extinction events, signals of mass extinction were still detectable on the resulting maximum clade credibility (MCC) trees. The accuracy of the detection was similar to the one obtained from MCC trees inferred using an FBD model that includes mass extinction parameters. We also reduced the fossilization rate and the number of morphological characters, obtaining results consistent with the aforementioned findings. However, reducing the fossilization rate decreased the accuracy of detecting mass extinctions when they occurred, and reducing the number of morphological characters decreased the accuracy of divergence time inference. Furthermore, we adjusted the priors for the existence of mass extinction and the survival probability of mass extinction. We found that the prior for the existence of mass extinction had no effect on inference, whereas the prior for the survival probability of mass extinction significantly influenced both the detection of mass extinctions and the estimation of survival probabilities. It should be noted that our simulations represent a largely best-case scenario with constant diversification rates; empirical data sets with more complex evolutionary histories may present additional challenges for mass extinction detection. Finally, we applied these models to an empirical data set of crinoids and found that, consistent with our simulation results, the inclusion of a mass extinction event in the tree prior had a negligible impact on the inferred topologies and divergence times at the whole-tree level, although small variations on individual node estimates were observed in the empirical analyses.