Most of our knowledge about online news consumption comes from survey-based news market reports, partial usage data from a single editor, or what people publicly share on social networks. This paper complements these sources by presenting the first holistic study of visits across online news outlets that a population uses to read news. We monitor the entire network traffic generated by Internet users in four locations in Italy. Together these users generated 80 million visits to 5.4 million news articles in about one year and a half. This unique view allows us to evaluate how usage data complements existing data sources. We find for instance that only 16% of news visits in our datasets came from online social networks. In addition, the popularity of news categories when considering all visits is quite different from the one when considering only news discovered on social media, or visits to a single major news outlet. Interestingly, a substantial mismatch emerges between self-reported news-category preferences (as measured by Reuters Institute in the same year and same country) and their actual popularity in terms of visits in our datasets. In particular, unlike self-reported preferences expressed by users in surveys that put “Politics”, “Science” and “International” as the most appreciated categories, “Tragedies and Weird news”’ and “Sport” are by far the most visited. We discuss two possible causes of this mismatch and conjecture that the most plausible reason is the disassociation that may occur between individuals’ cognitive values and their cue-triggered attraction.