Objective: This pilot, exploratory study explores the feasibility and usability of an AI-driven system (ChatGPT-4.0) as an accessible virtual coach for home-based recreational yoga. Materials and Methods: Two ChatGPT-generated documents were produced: (i) a 30-minute baseline sequence and (ii) an expanded version with pose instructions. Data sources comprised 11 beginner participants who attempted the sequences at home-practice level, 3 certified yoga experts who performed a written-instruction review, and researcher observations. Feasibility/clarity was rated per pose with a 0/0.5/1 rubric (0=not understood/applicable; 1=clearly understood/fully applicable). Results: Across 12 items, 9/12 poses achieved ≥80% feasibility (mean score ≥0.80 out of 1.00). Simple/static postures were consistently high: Seated Side Stretch, Adho Mukha Svanasana, Baddha Konasana, Paschimottanasana, and Savasana each scored 100% (15/15); Breathwork 96.7% (14.5/15); Marjaryasana–Bitilasana and Supta Matsyendrasana 86.7% (13/15); Shoulder–Neck Stretching 80.0% (12/15). In contrast, dynamic/complex items were lower: Surya Namaskar A 56.7% (8.5/15), Anjaneyasana 60.0% (9/15), and Ardha Kapotasana 10.0% (1.5/15). The absence of real-time supervision surfaced as a safety limitation for complex transitions. Conclusion: In a recreational, home-based context, ChatGPT shows promising feasibility for foundational poses and can function as a supplementary tool that supports self-directed practice. However, complex sequences require professional oversight, highlighting the need for multimodal feedback (e.g., audio/visual or sensor-based) in future developments.
Halime Dinç (Sat,) studied this question.