This paper explores the field of visual question answering applications based on remote sensing, focusing on the systems that use satellite imagery to generate textual outputs. As the volume of the geospatial data continues to grow, there is an ever-increasing demand for effective methods to interpret this information. This study combines recent tech advancements in the machine learning field and natural language processing techniques that enables the automatic generation of textual descriptions from the aerial images. This research work assesses the effectiveness of various existing Vision Language Models (VLMs) and their accuracy on distinct remote sensing tasks. This paper discusses the role of datasets, emphasizing the importance of the labelled training sets in enhancing the mode’s evaluation and performance. There are many challenges related to the ambiguity of interpretation of the images, the complexity of the language generation, and the need for domain-specific knowledge. By providing an overview of the current state of the VLMs in remote sensing, this study aims to inform the researchers and the practitioners about the capabilities, limitations, and the future prospects of these new innovative applications.