'Why Won't It Just Tell Me?': Students' Perceptions of a Reasoning-Support Chatbot in Education
Abstract
Generative AI chatbots are increasingly entering classrooms, yet by the time students encounter them in formal instruction, many have already used them outside school as tools for obtaining quick answers to school-related tasks. This creates a design challenge for educational chatbots intended to scaffold reasoning rather than provide ready-made answers. This poster presents an exploratory qualitative analysis of students' perceptions of Tiny, a teacher-monitored Socratic chatbot used in lower-secondary digital literacy lessons. Data consist of 11 semi-structured focus group interviews with approximately 80 eighth-grade students (age ~13--14 years) from four schools. Students interacted with Tiny during four lessons on facial recognition, servers, information manipulation, and recommender algorithms. We are conducting an inductive analysis of Czech transcripts, focusing on perceived usefulness, trustworthiness, and limitations. Preliminary findings suggest that students interpreted Tiny through prior expectations of AI as an answer-providing tool. Some students valued being pushed to think, but often disliked repeated questioning. Trust depended on conversational responsiveness: students became frustrated when Tiny repeated questions, misunderstood their input, or failed to help when they were stuck. The findings suggest that educational chatbots must be responsive, not merely Socratic, especially when they deliberately resist students' expectations of immediate answers.