Genome-scale metabolic network reconstructions (GENREs) are a powerful computational tool for mathematically modeling the metabolic processes within a cell at a systems-level. The development of improved curation methods through strategic data integration would improve our ability to use GENREs to understand metabolic diseases and to inform metabolic engineering. Metabolomics aims to identify metabolites within a biological system, which can then be integrated into a GENRE to increase its accuracy. Due to the cost of gathering metabolomics data, there is a need to identify which media conditions would hold the most value for model curation. To this end, we created an ensemble of draft GENREs for E. coli K-12 using a combination of well-established packages and in vitro anaerobic single-carbon source utilization screen data. Production sub-networks were created using weighted parsimonious flux balance analysis with different objective functions based upon single products across 44 candidate minimal media conditions with varied carbon sources. The average likelihood of reactions in each production sub-network was used as data to assess which ensembles had the greatest variation in network structure as a result of a given media condition. We were able to identify the 10 media conditions that induced the greatest variation among ensemble members, representing the conditions for which gathering metabolomics data would be most informative. This study developed a process for creating a prioritized list of media conditions for which to gather metabolomics data, in order to best increase the accuracy of a GENRE’s predictions and reduce the uncertainty in network structure.