Data availability remains a principal factor limiting the use of species distribution models (SDMs) as tools for wildlife conservation and management of rare species. Although data collected in systematic and rigorous fashion are preferable, available data for most species of conservation interest are usually low in both quality and number. Here we show that combining records published in peer-reviewed journals and gray literature sources (e.g., theses, government, and NGO reports) with unpublished records obtained by personal communications from relevant stakeholders affect the predicted distribution of spectacled bears (Tremarctos ornatus) in Peru. We built SDMs using generalized linear models, random forest, and Maxent, first using a dataset that only included published records, and second with a dataset using both published and unpublished records. All models were replicated ten times with random subsets with controlled sample size. Models that combined published and unpublished spectacled bear records had a better performance, irrespective of with SDM method used, increasing the connectivity of the species’ range, and increasing the overall predicted distribution area than models that only included published records. This was because unpublished records added key new localities, reducing spatial sampling biases. Our study shows that the inclusion of commonly disregarded data such as opportunistic records, reports from natural park rangers, student theses, and data-deficient small studies can make an important contribution to the overall ecological knowledge of rare and difficult-to-study species such as the spectacled bear.