This paper presents a novel methodology for assessing the similarity of local optimization algorithms generated by Large Language Models (LLMs), with the goal of distinguishing newly generated algorithms from traditional or previously used methods. In AI-based memetic algorithms, local search procedures are dynamically generated by LLMs based on characteristics of the optimization problem and contextual information from the optimization process, such as iteration number, applied global and local methods, and performance metrics. However, determining whether a newly generated algorithm is genuinely distinct or exhibits meaningful novelty compared to existing methods, remains a key challenge. The proposed assessment framework addresses this by integrating semantic and structural similarity analysis. We apply the framework to four LLM-generated local optimization algorithms and compare them with several classical gradient-based and gradient-free methods. In addition, numerical experiments are conducted using multi-dimensional benchmark functions to evaluate optimization performance. All results, including similarity rankings and performance comparisons, are presented and discussed.