In this paper, we provide empirical evidence for Tönnis' (2021) hypothesis that German cleft sentences address relatively unexpected questions in discourse while their canonical variants address relatively expected questions. We present an experiment that measures the relative preference between the German cleft and its canonical variant in contexts that differ with respect to how expected the question is that they answer. The expectedness of the question was measured separately in a norming study. The result of the experiment supports analyses of German clefts that take discourse expectations into account when analyzing the acceptability of clefts in contrast to canonical sentences. Approaches that primarily focus on differences in exhaustivity (e.g., De Veaugh-Geiss et al. 2018) or contrast (e.g., Rochemont 1986) need to be adapted in order to account for the results.