Global declines in bee populations are threatening the ecosystem services they provide, including pollination. Many bee–plant interactions are understudied, producing an incomplete understanding of resulting ecosystem-level vulnerabilities. The last decade has generated a wealth of opportunistic data originating from natural history collection records, published ecological datasets, and citizen/community science initiatives in online databases such as Global Biotic Interactions (GloBI). Here, we explore hypotheses related to bee–plant interactions and detection processes using the GloBI database, curated checklists of bee and flowering plant species, and an occupancy model. We hypothesized that larger, social bees would visit a larger number of plant species, while smaller, solitary bees would visit fewer. We also predicted that flowers with open, bowl-like shapes would attract a greater diversity of bee visitors compared to closed shapes. Further, we hypothesized that both floral and bee traits, such as bright colors and conspicuous patterns, would increase detectability, and that different data collection methods would vary in their ability to capture bee–plant interactions. Lastly, we hypothesized that the interaction network generated by the output of the occupancy model, which accounted for imperfect bee–plant detection, would yield more interactions, thereby increasing measures of evenness and decreasing nestedness and specialization, as compared to the network generated from recorded interaction data. We found that smaller bees exhibited higher probabilities of plant interactions than larger bees, but we did not find evidence that bee sociality influenced the probability of interacting with plants. We found that blue flowers and closed (not-bowl-shaped) flowers had higher probabilities of bee-plant interaction than other flower colors or bowl-shaped flowers, respectively. We also found that larger bee size, blue flowers, bowl shapes, and community science sources were associated with higher detection probabilities of bee–plant interactions. Lastly, the interaction network generated by the occupancy model output showed higher levels of evenness, nestedness, and connectance than the network generated by the GloBI data. Our study is among the first to utilize occupancy modeling to directly model species' interactions, leverage aggregated, open-source databases and expert checklists, and highlight the influence of detection and collection biases on our understanding of ecological interactions.