Although computational approaches to ideology-based audience fragmentation are promising, they are not without limitations. First, most existing studies have focused on the US, and the cross-national validity of their results has rarely been tested. Second, previous studies that rely solely on behavioral data from social media tend to make strong analytical assumptions such as that Twitter users prefer to follow media and politicians whose ideological positions are similar to their own, and that the ideologies of political elites can be extrapolated to infer the ideologies of ordinary users. We aim to address these limitations. First, we focus on Japan to test the generalizability of US findings in an Asian context. Second, we do not rely solely on behavioral measurement but combine survey and social media data to infer the ideologies of Twitter users. Results indicate that our classifier built based on these self-reported ideologies produces more valid estimates than naïve extrapolation from views of the elites. Based on these improved estimates of Twitter users’ ideologies, we demonstrate that the ideological distributions of those who follow major Japanese media accounts on Twitter largely overlap, suggesting an absence of ideology-based audience fragmentation.