“Talking with technology”: Conversation analysis of autistic student-chatbot interactions in SCHOOL KONTAKT™

Purpose Large language model chatbots can potentially augment manualised social skills programs aiming to support autistic adolescents, providing them with opportunities to engage in interactive discussions. The aim of the current study was to analyse conversations between autistic students and a custom chatbot developed for use with the SCHOOL KONTAKT™ social skills program. Method The study analysed 143 conversations between 31 autistic students and a chatbot based on Anthropic's Claude 3.5 and 4.5 Sonnet models. The structure of conversations was analysed using conversation analysis, providing insights on the methods used to engage students. Secondly, conversations were examined to establish whether they met the basic psychological needs outlined by self-determination theory. Finally, conversations were analysed using the Intelligence Systems Technology Acceptance Model to assess chatbot acceptance. Results Chatbot responses were positive and supportive, although often lengthy, sometimes overwhelming students. They were indirect, making little use of sarcasm, idioms, or analogies. Students were direct in responding to the chatbot. The chatbot supported autonomy, providing students with control over request execution and offering praise, supporting competence. Some students found the chatbot difficult to use. Students offered personal information to the chatbot, demonstrating trust. Minimal concern for transparency was demonstrated in conversations. Conclusion The chatbot offered autistic students a safe, accessible space for working through SCHOOL KONTAKT™. Future research might investigate adjusting system constraints to reduce complexity and promote ease of use. Additionally, increasing customisation would likely be welcomed, allowing students to personalise their experience.