An ensemble of forecasts is necessary to identify the uncertainty in predicting a non-linear system like climate. While ensemble averages are often used to represent the mean state and diagnose physical mechanisms, they can lead to information loss and inaccurate assessment of the model’s characteristics. We highlight an intriguing case in the seasonal hindcasts of the Climate Forecast System version-2. While all ensemble members often agree on the sign of predicted El Nino Southern Oscillation (ENSO) for a particular season, non-ENSO climate forcings, although present in individual members, are disparate. As a result, an ensemble mean retains ENSO anomalies while diminishing non-ENSO signals. This difference between ENSO and non-ENSO predictions and a more decisive impact of ENSO on seasonal climate increases the ensemble mean ENSO-Indian Summer Monsoon Rainfall correlation. Thus, a model’s teleconnection skills, which often help interpret physical mechanisms, should be studied using individual members rather than ensemble averages.