Today’s leaders need to have an adequate understanding of leadership for the digital age. This understanding we call “digital leadership”. It is a strategic knowledge asset and is indispensable for companies to be successful in a world characterised by volatility, uncertainty, complexity and ambiguity (VUCA). In their study Jäckli & Meier (2020) first asked the question as to what dimensions digital leadership actually consists of. After answering the question by establishing 10 dimensions, they next conducted surveys in 2018 and 2020 in Swiss companies to examine where they stand concerning digital leadership. It was planned to do the survey every two years. Conducting such surveys is time-consuming, mainly due to the fact that voluntary participants at the appropriate management level have to be identified, contacted and convinced to participate every second year. Additionally, experience shows that despite efforts, the response rate overall is low. Therefore, this study investigates the machine learning approach of text mining as an alternative option to the traditional survey. The conducted text mining-experiment is based on the data from the surveys of 2018 and 2020, using the contributors’ information contained therein as a basis for web scraping and training an artificial intelligence (AI) supervised learning model. The corpus obtained includes texts from 211 company websites, which are processed with Natural Language Processing algorithms and used for model training after labelling. This allows predictions to be made about digital leadership dimensions based on company websites. The experiment developed in the programming language Python showed that there is predictive power in company websites, but the prediction accuracy is low, ranging from 35 to 71 percent (49 percent on average) across all 10 digital leadership dimensions. Further, the imbalanced training data across all 5 values of the applied Likert scale leads to additional challenges. Predicting values that are rarely or not at all present in the training dataset is made impossible by the bias of the model. Opportunities exist to improve prediction accuracy with additional training data. Further surveys will most likely not correct the imbalance in the training data, but qualitative approaches in combination with unsupervised learning algorithms seem to be promising developments.