Abstract This study examines impacts on prediction skill of empirically‐derived tendency corrections (TC) to climatological seasonal cycle biases in retrospective seasonal forecasts from the Canadian Earth System Model version 5 (CanESM5). A novel aspect is that TC are applied simultaneously in the modeled atmosphere and ocean, rather than to either independently. As in previous studies, climatological seasonal cycle biases and associated model drift are substantially reduced in variables to which TC are applied. In addition, the skill of the predictions is appreciably improved, both on average globally and in particular regions such as the far western equatorial Pacific. These improvements extend to variables for which TC is not directly applied such as precipitation, and are in contrast to some previous studies in which TC resulted in at best minor improvements to skill.