We study the drivers of like-minded and low-reliability news-following on social media, as well as the effectiveness of intervention targeting them. We posit that, besides preferences for like-minded news and incomplete information about reliability, social media users might follow like-minded and low-reliability outlets because of platform cues that disproportionately surface pro-attitudinal and low-reliability content, thus making that content particularly salient in users' minds. We test the importance of these salience-driven behavioral frictions in a five-week field experiment with more than 3,000 U.S. Facebook users. Guided a theoretical framework, the experiment varies: (i) whether participants are prompted to re-optimize the portfolio of news pages they follow on Facebook through a platform-integrated interface that increases the salience of a balanced set of news pages, and (ii) whether they receive personalized information about outlet slant and reliability. We find that, consistent with our salience model and in contrast to canonical models of news consumption, the re-optimization interface with a salient menu of news pages induces large portfolio changes even without the provision of information, and that the provision of reliability information that users need but lack affects behavior only if paired with the re-optimization interface. Our interventions produce two main implications for users' news portfolios. First, they move users’ portfolios closer to their stated preferences, mitigating internalities. Second, they reduce the slant and increase the reliability of users' portfolios, thus potentially mitigating negative externalities for democracy. The induced portfolio changes persist for more than a month, translate into measurable changes in online news consumption, and are unlikely to be driven by experimenter demand effects.