This paper presents ReflectiveSigns, i.e. digital signage (public electronic displays) that automatically learns the audience preferences for certain content in different contexts and presents content accordingly. Initially, content (videos, images and news) are presented in a random manner. Using cameras installed on the signs, the system observes the audience and detects if someone is watching the content (via face detection). The anonymous view time duration is then stored in a central database, together with date, time and sign location. When scheduling content, the signs calculate the expected view time for each content type depending on sign location and time using a Naive Bayes classifier. Content is then selected randomly, with the probability for each content weighted by the expected view time. The system has been deployed for two months on four digital signs in a university setting using semi-realistic content & content types. We present a first evaluation of this approach that concentrates on major effects and results from interviews with 15 users.