Seasonal forecast models are known to suffer from systematic errors in their representation of El Niño-Southern Oscillation (ENSO) events. These errors include ENSO-related sea surface temperature (SST) anomalies that extend too far to the west in the tropical Pacific, and ENSO events that decay too slowly, particularly in the eastern Pacific. We have earlier shown that these errors develop rapidly, within weeks following forecast initialisation. In this study, we examine the early development of these errors in the European Centre for Medium-Range Weather Forecasts (ECMWF) seasonal forecast model version 5 (SEAS5). Using both a fully coupled hindcast dataset, and an uncoupled hindcast experiment forced by observed time-evolving SSTs, we evaluate the development of ENSO-related errors; that is, systematic errors as a function of forecast lead time, determined by regressing forecast error on the observed Niño3.4 index at time of initialisation. We confirm that atmospheric errors develop rapidly after initialisation, particularly errors in surface winds, wind stress and turbulent heat flux, in both the coupled and uncoupled hindcasts. These errors are similar in both datasets for the first two weeks, suggesting that the initial errors have an atmospheric origin. Subsequently, however, the errors diverge between the coupled and uncoupled hindcasts, demonstrating the importance of ocean-atmosphere coupling in their evolution. Coupled errors develop after approximately two weeks, including in SST and sea surface height, as the forecast error slowly evolves towards the model's own preferred state. In the uncoupled hindcasts, where the oceanic state is fixed to observations, the model is kept out of balance, meaning that in some cases the atmospheric errors are larger in the uncoupled hindcasts than the coupled hindcasts. The processes involved in the development of the initial errors and the coupled feedback will be discussed, including initial oceanic equatorial Kelvin wave and off-equatorial Rossby wave responses in the forecast error fields.