Our recent research has shown that preparing to listen to a different voice in a ‘cocktail party’ reduces the attention “switch cost”, but only if the probability of a relevant voice switch is relatively low. The current study investigates whether this effect of switch probability on the switch cost is due to “phasic” reconfiguration initiated before the target voice is specified when a switch is likely (pre-cue reconfiguration, e.g. Monsell & Mizon, 2006). To examine this, we employed the voice switching paradigm pioneered by Koch et al. (2011) and manipulated the response-cue interval (RCI) to investigate whether the probability of a switch in the target voice influenced the switch cost only when the RCI allowed sufficient time for pre-cue reconfiguration. We found that a longer RCI did not increase the effect of switch probability on the switch cost – indeed, this effect was robust even at the short RCI when there was little time available for pre-cue reconfiguration. These findings challenge the pre-cue reconfiguration account of the effect of switch probability on the switch cost. The results are more consistent with tonic/sustained accounts, which posit that a high switch probability may lead to multiple task-sets (here: attentional templates) being held active in Working Memory (e.g., Dreisbach & Fröber, 2019), or to a lower “commitment” to the current task/attentional set, expressed computationally by a lower gain parameter of the task/attentional set activation function (e.g., Musslick & Cohen, 2021).