This paper provides a rational for the inconsistencies in evaluating the embodiment of Pedagogical Conversational Agents (PA) in empirical research. The leading hypothesis for the bulk of studies in this field is that providing embodied agents in educational interfaces and online learning systems will increase the sociability of these interfaces, and thus will have a positive impact on learning and motivation. However, both quantitative and qualitative reviews did not confirm this effect. This paper tackles the lack of the robustness in the evaluation of embodiment or embodied interaction in learning contexts and argues for a reframing of the question of effectiveness to address the problem of embodiment.