For many scientific applications, it is highly desirable to be able to compare metabolic models of closely related genomes. In this short report, we attempt to raise awareness to the fact that taking annotated genomes from public repositories and using them for metabolic model reconstructions is far from being trivial due to annotation inconsistencies. We are proposing a protocol for comparative analysis of metabolic models on closely related genomes, using fifteen strains of genus Brucella, which contains pathogens of both humans and livestock. This study lead to the identification and subsequent correction of inconsistent annotations in the SEED database, as well as the identification of 31 biochemical reactions that are common to Brucella, which are not originally identified by automated metabolic reconstructions. We are currently implementing this protocol for improving automated annotations within the SEED database and these improvements have been propagated into PATRIC, Model-SEED, KBase and RAST. This method is an enabling step for the future creation of consistent annotation systems and high-quality model reconstructions that will support in predicting accurate phenotypes such as pathogenicity, media requirements or type of respiration.