How do we learn to control attention? Is it better to decide what to learn and – if so- why are self-organized learning schedules better? To study control of attention, we assessed how participants acquire different attentional policies in a context-appropriate way. We use a Stroop-like task and manipulated the proportion of congruent-to-incongruent trials separately for two contexts. Then we compared groups of participants that could decide which context to complete next (Choice-group) against groups of participants who were assigned to contexts randomly (Random-group) or based on choices of participants in the choice-group (Yoked-group). We measured how susceptibility to conflict varied with the context-specific proportion congruency (CSPC) manipulation and found larger CSPC effects for the choice and yoked groups compared to random. This suggests that in the CSPC paradigm voluntary control over context scheduling is beneficial, but not because of voluntary control, but rather due to optimal scheduling of context-control associations.