The sensitivity of urban canopy air temperature (T_a) to anthropogenic heat flux (Q_AH) is known to vary with space and time, but the key factors controlling such spatiotemporal variabilities remain elusive. To quantify the contributions of different physical processes to the magnitude and variability of ∆T_a/∆Q_AH(where ∆ represents a change), we develop a forcing-feedback framework based on the energy budget of air within the urban canopy layer and apply it to diagnosing ∆T_a/∆Q_AH simulated by the Community Land Model Urban over the contiguous United States (CONUS). In summer, the median ∆T_a/∆Q_AH is around 0.01 K ( W m⁻² )⁻¹ over the CONUS. Besides the direct effect of Q_AH on T_a, there are important feedbacks through changes in the surface temperature, the atmosphere–canopy air heat conductance (c_a), and the surface–canopy air heat conductance. The positive and negative feedbacks nearly cancel each other out and ∆T_a/∆Q_AH is mostly controlled by the direct effect in summer. In winter, ∆T_a/∆Q_AH becomes stronger, with the median value increased by about 20% due to weakened negative feedback associated with c_a. The spatial and temporal (both seasonal and diurnal) variability of ∆T_a/∆Q_AH as well as the nonlinear response of ∆T_a to ∆Q_AH are strongly related to the variability of ca, highlighting the importance of correctly parameterizing convective heat transfer in urban canopy models.