For effective guidance of agents’ opinions in social networks, it is important to understand how messages evolve and analyze their impact on agents’ opinions. Opinion dynamics model how agents influence each other’s opinions and how the entire network’s opinions evolve. In the literature, many works have used opinion dynamics to study the influence of messages on agents’ opinions. However, most works assume static messages or independence among messages at different times. Studies in mass media theory show that the message evolution process exhibits temporal continuity, randomness, and polarization features. In this work, we first propose the Bounded Brownian Message (BBM) model to describe the message evolution process, jointly considering the above features. We then combine the BBM model with the classic DeGroot opinion dynamics model and propose the Message EvoLution and Opinion DYnamics (MELODY) model to study the impact of message evolution on opinion dynamics. We theoretically analyze the probability distributions and statistics of messages and opinions and study how messages influence the agents’ steady-state opinions. Simulations and real user tests validate our analyses. This study is critical to a better understanding of how messages shape agents’ opinions in social networks and design effective mechanisms to guide agents’ opinions.