Conversational Agents (CAs) such as ChatGPT have been increasingly implemented in educational contexts to support teaching, learning, and administrative tasks, particularly in higher education. However, current CAs often lack personalized conversational capabilities that could enhance learning support. Embodiment plays a crucial role in improving these interactions, as Embodied Conversational Agents (ECAs) are virtual characters that can communicate with users in a lifelike manner through natural language and non-verbal cues, including facial expressions and gestures. Despite their potential, ECAs remain under-researched in relation to pupils. While previous research has examined the general role of ECAs in education, there is a significant gap concerning their design and effectiveness for pupils.

To address this gap, we conducted a workshop involving 64 pupils' aged 13 to 16 to investigate how ECAs should be designed to optimally support pupils' in this age group. Our workshop aimed to identify the tasks for which ECAs are most effective and to determine the key properties that contribute to their educational value and user engagement. To gain deeper insights, we designed and conducted a structured workshop in which we systematically collected data and gathered pupils' perspectives on their experiences and preferences regarding ECAs.

Our findings provide a deeper understanding of how ECAs can be effectively integrated into educational settings for pupils. The results contribute to the development of more engaging and supportive learning environments by informing the design of ECAs that align with pupils' expectations and preferences. Future research should further explore the impact of age-related differences on pupils' preferences for ECAs.

Keywords: Embodied Conversational Agents (ECAs), Education, Educational ECA, Personalized Learning.