Understanding what drives textual success is central to literary studies and computational linguistics. This study models fanfiction popularity using surprisal, a probabilistic measure of word predictability, as a processing-based proxy for readability. Using a dataset of fanfiction texts and a composite measure of reader engagement, results show a small negative association between mean surprisal and popularity. However the relationship follows an inverted U shape suggesting that moderate surprisal maximizes engagement. Lower surprisal variance is also linked to higher popularity consistent with the Uniform Information Density hypothesis. Additional analyses indicate variation across genres and fandoms. Overall successful texts appear to balance predictability and novelty.