Due to the complexity of consumer behavior, it is difficult to comprehensively capture and accurately analyze, resulting in significant push errors in practice and inability to achieve the expected push effect. Therefore, a research on adaptive push of agricultural product sales information based on composite neural networks is proposed. Using web crawler technology to crawl user social data information on sales websites, segmenting and stopping the crawled data information, extracting interest feature values, and combining CNN neural network and GRU recurrent neural network to more accurately capture and analyze changes in farmers’ interest in agricultural product sales information, providing strong support for personalized push. Using CNN neural network to extract the interest features of farmers in agricultural product sales information, and training the extracted features using GRU recurrent neural network to generate an adaptive push list of agricultural product sales information, achieving adaptive push of agricultural product sales information based on combination neural network. Experimental results have shown that the push error of the design method does not exceed 0.1, and it can achieve adaptive and accurate push of agricultural product sales information.