In organic photovoltaics, the role of hot charge-transfer (CT) excitons on free carrier generation is currently under intense debate. In this project, we carry out first-principles time-dependent density functional theory calculations to examine hot CT dissociation in polymer/fullerene heterojunctions. We reveal that whether or not hot CT states promote charge separation depends on excitation spectral range and crystallinity of the donor and acceptor phases. We find that while the crystallinity of the donor phase underlies the energy dependence of CT exciton dissociation, the crystallinity of the acceptor determines charge separation efficiency. We propose a theory that can reconcile contradictory experimental observations and provide insight into hot CT dissociation. Crucially, the timescale of hot CT dissociation is found to be comparable to the timescale of its relaxation to the lowest-lying CT state, which is localized in all interfacial models studied.