In this paper, we aimed to test whether we could predict sport type (badminton or running) and marathon proficiency from the valence, form, and content of the athletes' self-reported inner speech. Additionally, we wanted to assess the difference between self-talk during high intensity and low intensity exercise. The present study corroborated existing research - we were able to predict both sport type in Study 1 and intensity level as well as marathon proficiency in Study 2 from questionnaire data using machine learning models. In Study 1, we found that badminton players engage more in worry and anxiety-control while runners are more prone to task disengagement. Interestingly, it seemed in Study 2 that the more participants engaged in condensed, positive, and repetitive self-talk when not pushing themselves, the slower their fastest marathons and half marathons were. We discuss potential explanations for these findings and make suggestions for future research.