In the current context of global change and a biodiversity crisis, there are increasing demands for greater predictive power in ecology, in both the scientific literature and at the science–policy interface. The implicit assumption is that this will increase knowledge and, in turn, lead to better decision‐making. However, the justification for this assumption remains uncertain, not least because the definition of ‘prediction’ is unclear. We propose that two types of prediction should be distinguished: corroboratory‐prediction is linked to the validation of theories; and anticipatory‐prediction is linked to the description of possible futures. We then discuss four families of obstacles to prediction, linked to the specific features of ecosystems: 1) they are historical entities, 2) they are complex, 3) their dynamics are stochastic, and 4) they are influenced by socio‐economic drivers. A naïve understanding of ecological science suggests that the two types of predictions are simply two phases in a sequence in which scientists first improve their knowledge of ecological systems via corroboratory‐predictions, and then apply this knowledge in order to forecast future states of ecosystems via anticipatory‐predictions in order to help policy makers taking decisions. This sequence is however not straightforward, partly because corroboration and anticipation are not affected by the obstacles to prediction in the same way. We thus invite to reconsider the role of ecological prediction as a tool in a deliberative model of decision‐making rather than as external scientific information aimed at enlightening the political process. Doing so would be beneficial for both the policy‐relevance of anticipatory‐prediction and the theoretical‐relevance of corroboratory‐prediction.