This academic paper introduces Augmented Marginal outcome density Ratio (AMR), a novel approach for estimating average treatment effects (ATE) from observational data that addresses limitations of existing methods, particularly in settings with high-dimensional covariates and weak overlap. Unlike covariate-focused adjustment techniques prone to sensitivity in complex scenarios, AMR employs outcome-informed weighting to naturally filter irrelevant information and enhance robustness. The authors demonstrate that AMR is doubly robust and achieves asymptotic normality, while empirical results on synthetic and real-world datasets, including text applications, highlight its superior efficiency and stability compared to various baseline methods, especially under challenging conditions.