COMPARATIVE EVALUATION OF AI-GENERATED AND COACH-DESIGNED TRAINING PROGRAMS IN YOUTH KICKBOXING, ПОРІВНЯЛЬНА ОЦІНКА ТРЕНУВАЛЬНИХ ПРОГРАМ У ЮНАЦЬКОМУ КІКБОКСИНГУ, ЗГЕНЕРОВАНИХ ШТУЧНИМ ІНТЕЛЕКТОМ ТА РОЗРОБЛЕНИХ ТРЕНЕРАМИ

<jats:p>Introduction. The aim of this study was to conduct an expert evaluation and comparative analysis of kickboxing training programmes developed by coaches of different qualification levels and by an artificial intelligence language model (ChatGPT). Materials and Methods. A blinded repeated-measures design was applied. Sixteen independent experts in striking combat sports evaluated four 8-week training programmes for beginner kickboxers aged 10–11 years: three coach-designed programmes (TP-L1 – expert, TP-L2 – intermediate, TP-L3 – novice) and one AI-generated programme (TP-AI). Programme quality was assessed using a 12-item questionnaire covering three domains: structure, combat sports specificity, and practicality. All items were rated on a 5-point Likert scale. Data were analysed using non-parametric statistics (Friedman test, Wilcoxon signed-rank test with Bonferroni correction, Kendall’s W). Results. Significant differences were found between programmes (χ²(3) = 32.30, p &lt; 0.01) with a strong level of inter-expert agreement (W = 0.67). The highest overall ratings were observed for TP-L1 (4.08 ± 0.32), followed by TP-L2 (3.50 ± 0.42) and TP-AI (3.32 ± 0.48), while TP-L3 received the lowest scores (3.01 ± 0.48). Pairwise comparisons revealed significant differences between TP-L1 and all other programmes (p &lt; 0.05–0.01), as well as between TP-L2 and TP-L3 (p &lt; 0.05). No significant differences were found between TP-L2 and TP-AI (p = 0.18) or TP-L3 and TP-AI (p = 0.63). Effect sizes were large for all significant comparisons (r = 0.67–0.99). Domain-specific analysis showed that TP-L1 achieved the highest scores across all domains, while TP-AI demonstrated relatively high structural quality (3.90 ± 0.57) but lower combat sports specificity (2.65 ± 0.64). Item-level analysis confirmed that the AI-generated programme performed well in general structure (Q1–Q4) but showed weaker results in sport-specific components (Q6–Q8). Conclusions. AI-generated training programmes demonstrate acceptable methodological quality and may reach a level comparable to those developed by less experienced coaches. However, expert-designed programmes remain superior, particularly in sport-specific integration. Artificial intelligence can be considered a supportive tool in training design, but professional coaching expertise remains essential in combat sports.</jats:p>