The electrochemical CO 2 reduction reaction is crucial for reducing atmospheric CO 2 and achieving carbon neutrality. Recent researches have predominantly focused on the development of high-performance catalysts, evaluating their performance remains time-consuming. Concurrently, in situ spectroscopic techniques have been instrumental in elucidating catalytic mechanisms and fundamental reaction pathways. In this study, we employ constant potential ab initio molecular dynamics simulations to precisely model the adsorption behavior of CO 2 molecules on Cu-Ag alloy surfaces, with a particular focus on tracking variations in the bond angles of the adsorbed CO 2. By generating 2600 unique spectral datasets, we leverage convolutional neural networks to extract key spectral features and train a deep learning model to predict the bond angles of adsorbed CO 2 based on the corresponding spectral information. Given that the bond angle of adsorbed CO 2 serves as a crucial descriptor of a catalyst’s ability to activate CO 2 molecules, our approach demonstrates the efficacy of AI in predicting catalytic activity. Furthermore, we extend this model’s applicability to Cu-Au and Cu-Zn alloys, establishing the potential for AI-driven preliminary catalyst screening based on spectral data. This methodology has the capacity to significantly accelerate the catalyst development pipeline by reducing the reliance on analyzing conventional experimental results manually.