Bibliographic coupling, over the years, has been referred to and used in different contexts related to scientific and technical literature. It is often believed that research papers that have bibliographic coupling deal with similar concepts and hence there may be high conceptual similarity between them. This study attempts to empirically asses this notion. To conduct this research, the study utilizes the data obtained from the Dimensions database and employs advanced machine learning algorithms to extract weighted keywords that better capture the conceptual content of documents. The Jaccard similarity measure is used to compute bibliographic and conceptual coupling matrices for different sets of research papers. The results show that even though bibliographic coupling is widely used to assess relationships between research papers, it often falls short of identifying actual conceptual similarities within documents. This study’s findings carry important implications for areas such as information retrieval, interdisciplinary research and evaluation metrics, calling for a more refined understanding of how research documents relate to one another beyond their shared references.