Résumé : Promoting diverse and abundant flowering plants in cities is essential to counteract the decline in wild bee diversity due to urbanisation. To support effective conservation, it is crucial to identify which flowers best enhance wild bee abundance and species richness. This requires large datasets and robust analytical tools. Here, we use a plant selection tool developed by M’Gonigle et al. (2016), which recommends flower mixes that maximise pollinator species richness based on visitation data. We analysed bee-flower interaction data from the Brussels Capital Region (Belgium), comparing two contrasting sources: (1) citizen science records and (2) standardised academic surveys. We evaluated the bipartite networks of these datasets and their combination and generated optimised flower mixes from each using the plant selection tool. Our results show that dataset composition and inherent biases strongly influence outcomes. The bipartite networks differed substantially (compositional difference = 0.86), mainly due to rewiring of bee-flower interactions (0.69). Consequently, the flower mixes derived from each dataset overlapped by only 7% when optimising for species richness. The combined dataset network more closely resembled the citizen science data (WN = 0.106) than the academic survey data (WN = 0.590).These findings highlight the substantial impact of data collection methods on ecological recommendations. Awareness of such biases is essential for making sound, evidence-based conservation decisions. To support wider application, we developed a free app that allows users to create flower mixes optimized for pollinator abundance, species richness, or both, using our dataset or their own.