Wildfires have caused increasingly negative impacts with increasing occurrences close to densely populated regions. Evacuations are among the most critical measures in the immediate wildfire relief measures. While social media have been used in natural disasters, there has been limited understanding of the efficacy of using social media to aid evacuations. This paper presents a data-driven study of social media-aided evacuations for the 2020 wildfires in the western United States, based on 53,990 relevant tweets. First, we analyzed the aggregated social media data and validated its reliability against information from official channels. Both the temporal and spatial investigations show good agreements with official information. Further, we classified the tweets into pre- and on-evacuation based on extracted word patterns. The classifications align well with evacuation levels from official channels. Next, we demystified the information dissemination patterns via network analysis. We have found that government channels, news agencies, and public figures prevail among top users. The top users for on-evacuations tend to be more local-focused than pre-evacuations. This study demonstrates the efficacy of using social media to aid evacuations. In addition, it provides guidelines for future studies on extracting high-priority information from social media for disaster relief.