This study introduces a more advanced framework for testing the performance of language models in nutrition, routine, and workout, specifically their capability to answer complicated, context-dependent questions. The methodology employs a range of prompting techniques, including Zero-Shot, Few-Shot, Chain-of-Thought, and Ensemble/Self-Consistency, in order to evaluate the performance of large language models (LLMs) in multi-step reasoning tasks, contextual understanding, and knowledge of specific domains. These are based on real-life activities in health and fitness, focusing on the specifics of muscle activation, grip positioning, and tailored workout. Tasks are paired with a standardized dataset of four hundred well-curated prompts in order to lead the model into generating responses regarding specific queries within the scope of health, nutrition, and fitness routines. Through the use of both automated query methods and manual assessment of the results of these questions, both techniques are evaluated, and each is analyzed to conclude on the consistent reliability of output from either method. The performance of different prompting strategies is compared and their strengths and weaknesses are identified in the context of health and fitness applications. The research highlights the importance of combining multiple prompting techniques to achieve more accurate and insightful results, which contributes to a deeper understanding of how language models can be applied to solve real-world problems in health and wellness.