Answers are grounded in this paper and cite their sources.
Teachers increasingly consult large language model (LLM) chatbots, yet little is known about the feedback guidance these conversational interfaces give physical education (PE) teachers, or whether it depends on the chatbot or the language. We submitted one prompt, describing a low-skilled child who repeatedly fails a forward roll and a teacher tempted to reprimand him publicly, to ChatGPT, Gemini, Claude, and DeepSeek in English, French, and Arabic. In this exploratory single-run study, a directed content analysis of the 12 responses (109 scripted feedback statements) described and compared chatbots and languages. All chatbots rejected the public reprimand and prescribed a chin tuck and a simpler task. However, only half explained the cause of the error, two-thirds of the technical cues combined several points, 95% of the technical cues directed attention to the body, and six responses contradicted their own guidance to give one cue at a time. Praise was often unspecific, one statement invited the student’s view, and several responses flattered the teacher. Descriptively, the chatbot accounted for 58% of the dissimilarity between coded responses and the language for 12%, and these shares cannot be separated from run-to-run variation. Within these limits, teachers may wish to check diagnosis, cue load, and autonomy support.
No takes yet. Share an insight, caveat, or question.
Romdhani et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: