In this paper, I find novel evidence that the language of the CEO during earnings calls can predict upcoming leadership turnover. Specifically, in meetings preceding the public announcement of a CEO transition, executives exhibit changes in business tone, stress levels, and self-attribution. The results are obtained using a state-of-the-art financial language model, FinBERT, coupled with traditional word-counting algorithms. The findings suggest that qualitative information in earnings calls contains signals about CEO turnover risk that are incremental to commonly used quantitative performance variables.