Recent research suggests that infants and toddlers succeed at a wide range of non-elicited-response false-belief tasks (i.e., tasks that do not require children to answer a direct question about a mistaken agent's likely behaviour). However, one exception to this generalization comes from verbal anticipatory-looking tasks, which have produced inconsistent findings with toddlers. One possible explanation for these findings is that toddlers succeed when they correctly interpret the prompt as a self-addressed utterance (making the task a non-elicited-response task), but fail when they mistakenly interpret the prompt as a direct question (making the task an elicited-response task). Here, 2.5-year-old toddlers were tested in a verbal anticipatory-looking task that was designed to help them interpret the anticipatory prompt as a self-addressed utterance: the experimenter looked at the ceiling, chin in hand, during and after the prompt. Children gave evidence of false-belief understanding in this task, but failed when the experimenter looked at the child during and after the prompt. These results reinforce claims of robust continuity in early false-belief reasoning and provide additional support for the distinction between non-elicited- and elicited-response false-belief tasks. Three accounts of the discrepant results obtained with these tasks – and of early false-belief understanding more generally – are discussed.