As electric vehicles (EV) become increasingly popular, issues such as the limited number of charging stations, low utilization rates, and suboptimal placement have drawn significant attention. Therefore, the rational planning of charging station locations is of paramount importance. Traditional site selection methods often require high-quality data inputs and are prone to overfitting, resulting in poor generalization. This study innovatively proposes converting regional characteristics into natural language text and introduces the PETRoBERTa model based on prompt learning to assess the suitability of different areas for constructing charging stations. The study focuses on Wuhan, using hourly time granularity and kilometer spatial granularity to predict the suitability of different grids for station construction. The proposed model is compared with other baseline models, and the results show that the PETRoBERTa model achieves a prediction accuracy of 93.21%, outperforming others across various evaluation metrics. Therefore, our method can effectively aid in the planning of charging station layouts, making a significant contribution to the further adoption and promotion of electric vehicles.