Abstract
AI-based chatbots are increasingly used to design training programs, but their effectiveness for elite athletes is unclear. This study assessed ChatGPT-4's ability to generate weekly training plans for elite swimmers and sprinters. Twenty-three coaches and thirty-six athletes rated the AI-generated plans using a 5-point Likert scale in three areas: weekly frequency, intensity adjustments, and training structure. Seven intensity zones were analyzed: A1 (endurance/recovery), A2 (extensive aerobic), B1 (intensive aerobic), B2 (aerobic-anaerobic transition), C1 (anaerobic threshold), C2 (anaerobic-lactate), and C3 (maximal sprint intensity). Coaches gave neutral-to-positive ratings (3.6 for distance swimmers, 3.7 for sprinters), while athletes were more critical (2.8 and 3.1, respectively). AI-generated plans performed well in low-intensity zones (A1) but had shortcomings in moderate-intensity (A2, B1-B2: long repetitions, excessive sets, insufficient recovery) and anaerobic zones (C1: excessive frequency for swimmers; C2-C3: insufficient frequency for sprinters). No significant differences emerged between plans for swimmers and sprinters (p=0.596), but A2, B1, and B2 showed greater discrepancies (p < 0.001). Rating reliability was moderate for coaches (ICC=0.609) and low for athletes (ICC=0.369). Older coaches and male athletes rated the plans lower, while those with national-level experience were more favorable. While 65% of coaches found the plans usable with minor modifications, only 27.8% of athletes agreed, 47.2% requested major changes, and 25% rejected them. ChatGPT-4 is useful for simple training plans but requires human supervision for complex periodization, particularly in high-intensity zones.