Plants open their stomata to take in carbon dioxide during photosynthesis, losing water simultaneously through transpiration. If there is low water supply or high water demand, plants can close their stomata to avoid hydraulic failure, thus shutting down photosynthesis. Model misrepresentation of this plant response to water stress is a major source of uncertainty in our future climate projections. Here we use observational and reanalysis data of land surface and near surface air temperature to create a novel emergent constraint for canopy conductance sensitivity to moisture availability in Earth System Models. We find that models tend to underestimate this sensitivity, particularly in semi-arid regions such as grasslands, croplands, and savannas. Based on the strong relationship between stomatal conductance and gross primary production, this also translates to a model overestimation of carbon uptake in these regions during dry anomalies, which we verified using flux tower data. Our results provide a novel benchmark to improve model representation of vegetation dynamics in these regions, thus improving forecasting ability for carbon, water and energy fluxes under future climate change scenarios.