Conditional cooperation is usually investigated in experiments where others’ choices are known. In this study, we explore conditional cooperation under uncertainty. Using a novel experimental design, we exogenously manipulate the likelihood that a subject’s partner in a Prisoner’s Dilemma will cooperate. Information about the partner’s cooperation is either presented descriptively or learned through experiential sampling. We find a description-experience gap: subjects are more likely to cooperate under experience than description when the likelihood of their partner’s cooperation is low, while the opposite holds when it is at least 50%. This finding is contrary to expectations from individual choice literature, where rare events typically receive less weight in experiential-based decisions. Our findings indicate that conditional cooperators are less responsive to social information when obtained experientially rather than descriptively, and illustrate how stronger priors under social than under individual uncertainty can account for this disparity.