Does the sense of self-efficacy affect the modalities of discourse production? The participants in our experiment studied a scientific test and then explained its contents in three consecutive trials, each followed by a feedback (fictitious) concerning a student’s comprehension of the lesson. We assume that participants, in order to study the text, build a mental model and exploit the model to organize their explanation. Further, we assume that positive feedbacks induce the participants to keep constant both the linguistic and the extralinguistic aspects of their explanation. Negative feedbacks, instead, induce a change in the way the model of the text is exploited, changes that result in an increasing use of diagrams, iconic gestures, and production of inferences with respect to the original text. The result of our experiment on 32 adult participants confirm our expectations.