While recycling facilities have been significantly upgraded in China, but the effectiveness of these facilities in improving waste management needs to be evaluated. Here, we conducted a nationwide survey by directly taking photographs of the inside of individual waste containers over 11 cities across China. We found that waste from recycling and non-recycling containers generally comprised similar materials. The corresponding waste features extracted by machine learning models tend to be well-mixed but clearly separated after removing the misplaced items, demonstrating an objective means for quantifying the accuracy of waste-sorting process. We therefore proposed the nationwide scale-up of this automated machine learning system, which along with additional incentive programs for better waste-sorting behaviors, may help improve waste management.