The linearity of global‐mean outgoing longwave radiation (OLR) with surface temperature is a basic assumption in climate dynamics. This linearity manifests in global climate models, which robustly produce a global‐mean longwave clear‐sky (LWCS) feedback of 1.9 W/m²/K, consistent with idealized single‐column models (Koll & Cronin, 2018, https//:doi.org/10.1073/pnas.1809868115). However, there is considerable spatial variability in the LWCS feedback, including negative values over tropical oceans (known as the “super‐greenhouse effect”) which are compensated for by larger values in the subtropics/extratropics. Therefore, it is unclear how the idealized single‐column results are relevant for the global‐mean LWCS feedback in comprehensive climate models. Here we show with a simple analytical theory and model output that the compensation of this spatial variability to produce a robust global‐mean feedback can be explained by two facts: (1) When conditioned upon free‐tropospheric column relative humidity (RH), the LWCS feedback is independent of RH, and (2) the global histogram of free‐tropospheric column RH is largely invariant under warming.