We develop learning-augmented algorithms for vertex-weighted and unweighted online bipartite fractional matching, with application to AdWords, and establish new robustness-consistency tradeoffs.