The high adoption of mobile phones coupled with 3G technology can extend Internet access to new communities. Such access, however, is impractical because mobile phone interfaces are cumbersome to use. In addition, hierarchical menus and search engines pose an interaction barrier to such communities. A content recommender is proposed to address these issues. Collaborative filtering is a technique that makes predictions regarding the preference of unobserved items based on the predictions of similar users. Unlike web-based implementations of these schemes where items can be explicitly rated, preference information in the mobile environment needs to be gathered purely implicitly. An evaluation is conducted into how quickly user-based collaborative filtering algorithms can identify preferred content based purely on user-content interactions. The evaluation of two similarity measures: Pearson correlation and vector similarity is conducted empirically in Matlab with the MovieLens dataset and are compared against a scheme that randomly recommends items. Vector similarity is observed to outperform Pearson correlation in certain cases. Results suggest that prior data regarding the user's preferences is required to reliably recommend content.