With the development of information and communication technologies, policy-makers and researchers focus on improving information empowerment to reduce poverty effectively, but there is still no consensus on the mechanism of information empowerment to promote poverty reduction. Based on panel data covering 124 countries (regions) from 1990 to 2017, this paper utilizes random forest in machine learning to select control variables and employs empirical models to demonstrate that information empowerment has a positive effect on reducing the poverty incidence among countries (regions), but the influence represents the non-linear characteristic with two critical inflection points. The further analysis interprets that there are significant differences between the long-term and short-term dynamic effects. At the same time, the influence involves two aspects: information empowerment improves external context causing poverty, thus forming an exogenous stimulating mechanism for poverty reduction; and it has path-dependence on endogenous context during alleviating poverty.Robustness tests such as instrumental variables confirm the consistency of those conclusions mentioned above. This study can provide a theoretical and practical supports for late-developing countries to get ahead in development by information empowerment, and a new perspective for China to advance theories about poverty alleviation. Finally, the paper discusses the theoretical, methodological and policy implications.