Artificial Intelligence (AI) is increasingly used in medical research, especially in the analysis and interpretation of medical data. In this study, based on the traditional CARS model, we built a framework for thesis abstract language step research by categorizing fuzzy steps into optional steps and adding appropriate key steps to the language steps. With the help of artificial intelligence technology, an extraction model of key elements of abstracts incorporating the attention mechanism is constructed, aiming at screening the elemental utterances in abstracts. Finally, by collecting data from medical related papers in CNKI, Web of Science and other databases, the CARS modeling strategy based on artificial intelligence is implemented in the comparative analysis of medical paper abstracts in English and Chinese. Through the comparative analysis, it is found that the number of sentences in English abstracts is concentrated in 6-7 sentences, while the number of sentences in Chinese abstracts is scattered in 2-8 sentences. The percentage of the use of Chinese sentences on English abstract writing is the highest, with an average percentage of 45.24%. The frequency of the first 20 words of fuzzy restrictive phrases in English abstracts was significantly higher than that in Chinese abstracts. The organization of Chinese and English abstracts was mostly in the structure of “introduction method-results-discussion”, which accounted for 54% and 71%, respectively. In addition, the frequency of steps indicating gaps in the research area is higher in English than Chinese abstracts.