Abstract Fossils provide our only direct window into evolutionary events in the distant past. Incorporating them into phylogenetic hypotheses of living clades can help elucidate macroevolutionary patterns and processes, such as ancestral states and diversification dynamics. However, the effect fossils have on phylogenetic inference from morphological data remains controversial. Previous studies have highlighted their strong impact on topologies inferred from empirical data, but have not demonstrated that they improve accuracy. The consequences of explicitly incorporating the stratigraphic ages of fossils using tip-dated inference are also unclear. Here we employ a simulation approach to explore how fossil sampling and missing data affect tree reconstruction across a range of inference methods. Our results show that fossil taxa improve phylogenetic analysis of morphological datasets, even when highly fragmentary. Irrespective of inference method, fossils improve the accuracy of phylogenies and increase the number of resolved nodes. They also induce the collapse of ancient and highly uncertain relationships that tend to be incorrectly resolved when sampling only extant taxa. Furthermore, tip-dated analyses which simultaneously infer tree topology and divergence times outperform all other methods of inference, demonstrating that the stratigraphic ages of fossils contain vital phylogenetic information. Fossils help to extract true phylogenetic signals from morphology, an effect that is mediated by both their unique morphology and their temporal information, and their incorporation in total-evidence phylogenetics is necessary to faithfully reconstruct evolutionary history.