The impact of artificial intelligence (AI) on the labor market, based on the Routine-Biased Technological Change paradigm, is widely acknowledged. However, existing job classification methods lack detail and accuracy. To address this limitation, the Chinese-BERT-wwm model was optimized to classify recruitment data from listed companies between 2013 and 2019 into routine and non-routine jobs, achieving a test set accuracy of accuracy of nearly 93%. Additionally, the GLM4 model was used to match job titles and descriptions to the "Chinese Occupational Classification (2022 Edition)" to identify digital occupations and analyze the impact of AI technology on labor demand structure. Empirical results show that higher AI technology levels significantly increase demand for non-routine jobs and reduce demand for routine jobs, with pronounced effects in non-state-owned enterprises, high-tech industries, and manufacturing. Further analysis reveals that the increased demand for non-routine jobs is primarily driven by growth in non-routine cognitive positions. Mechanism analysis shows that AI adoption increases non-routine job demand through productivity effects and the creation of new digital occupations, while reducing routine job demand through substitution effects. It expands the application of large language models in economic text analysis.