Ecological network theory hypothesizes a link between structure and stability, but this has mainly been investigated in-silico. In an experimental manipulation, we sequentially removed four generalist plants from real plant-pollinator networks and explored the effects on, and drivers of, species and interaction extinctions, network structure and interaction rewiring. Cumulative species and interaction extinctions increased linearly with removing plants, and both species and interactions disappeared faster than expected by co-extinction models, which even predicted several false cases. Networks were not stable and symptoms of fragility emerged with removing plants: nestedness decreased, modularity increased, and opportunistic random interactions and structural unpredictability emerged. Conversely, interactions reorganization (“rewiring”) was high, asymmetries between network levels emerged as plants increased their centrality and no change was found in stochastic robustness index. Our study shows that experimental manipulations of real networks indicate how species and interaction occurrences are altered when key resources are removed from the system.