<p>The rise of social media has fueled the propagation of fake news, which can distort public perception and consequently impact political decision-making and social stability. Current fake news detection methods predominantly focus on linguistic information analysis and dissemination structure modeling, yet they overlook the propagation dynamics features of information diffusion. Additionally, acquiring real-world propagation dynamic features requires complete propagation chain data. Therefore, we propose a propagation dynamics-derived fake news detection framework (called ProFNSE) that integrates the dynamic simulation of propagation dynamics features with linguistic features. Practically, to address the reliance of information propagation dynamics on complete propagation chains, we propose a dynamics analysis method leveraging large language models (LLM) simulation of propagation. By dynamically modeling the heterogeneity of user group backgrounds, the interaction mechanisms of viewpoints, and the differences in semantic sensitivity, our method enables effective extraction of propagation dynamic features. Additionally, the ProFNSE framework combines the general and latent features of textual information and employs multiple classifiers for training. Our experimental results on cross-lingual Chinese–English datasets show that the eXtreme Gradient Boosting (XGB) classifier achieved the highest accuracy of 0.8880 on the Chinese dataset, and the logistic regression (LR) classifier achieved the highest accuracy of 0.9717 on the English dataset.</p>