This study analyzes the relationship between financial news narratives and US government bond market volatility, as measured by the MOVE index. To do so, we analyze an underexplored dataset—the Global Database of Events, Language, and Tone (GDELT)—in which news articles published online are assigned metadata, such as topics and sentiment. Using this dataset, we propose the use of large language models (LLMs) to preselect topics for analysis, and we apply two techniques to identify those that contribute the most to the evolution of the MOVE index. First, we propose a LASSO algorithm that provides information on which topics covered in the news are most influencing the index's movement. Second, to reduce multicollinearity problems, we estimate a linear regression on previously identified clusters. Both methods are tested on three periods of high volatility in the MOVE index, starting in 2017. The results demonstrate that news narratives are significantly correlated with bond market volatility, and that the LASSO algorithm effectively identifies the most impactful narratives. This study provides valuable insights for investors and policymakers by linking financial news to bond market volatility, and paves the way for future research on the impact of these news stories on financial markets.