Abstract Background Phylogenomic approaches have great power to reconstruct evolutionary histories, however they rely on multi-step processes in which each stage has the potential to affect the accuracy of the final result. Many studies have empirically tested and established methodology for resolving robust phylogenies, including selecting appropriate evolutionary models, identifying orthologs, or isolating partitions with strong phylogenetic signal. However, few have investigated errors that may be initiated at earlier stages of the analysis. Biases introduced during the generation of the phylogenomic dataset itself could produce downstream effects on analyses of evolutionary history. Transcriptomes are widely used in phylogenomics studies, though there is little understanding of how a poor-quality assembly of these datasets could impact the accuracy of phylogenomic hypotheses. Here we examined how transcriptome assembly quality affects phylogenomic inferences by creating independent datasets from the same input data representing high-quality and low-quality transcriptome assembly outcomes. Results By studying the performance of phylogenomic datasets derived from alternative high- and low-quality assembly inputs in a controlled experiment, we show that high-quality transcriptomes produce richer phylogenomic datasets with a greater number of unique partitions than low-quality assemblies. High-quality assemblies also give rise to partitions that have lower alignment ambiguity and less compositional bias. In addition, high-quality partitions hold stronger phylogenetic signal than their low-quality transcriptome assembly counterparts in both concatenation- and coalescent-based analyses. Conclusions Our findings demonstrate the importance of transcriptome assembly quality in phylogenomic analyses and suggest that a portion of the uncertainty observed in such studies could be alleviated at the assembly stage.