With the continuous improvement of algorithmic recommendation technology and the rapid development of online media, network users have shown characteristics such as “crowdsourcing”, which has led to an increasing proportion of news and information dissemination methods based on content intelligent recommendation algorithms. This article explores the process of embedding artificial intelligence (AI) into intelligent recommendation algorithms for new media content, including collaborative filtering, rating matrix, matrix decomposition, feature embedding, recall, and policy ranking. According to the type of user behavior, positive feedback is categorized as playing, browsing, commenting, following, forwarding, etc. If there is no behavior, it is 0, and negative feedback is given to those who do not like it. In the process of matrix model decomposition, this article uses metadata for information hiding, reflecting the positive energy weight of mainstream media attributes on content. This article takes Today’s Headlines as an example to explore. After adopting the algorithm, the click through rate in the 6th month increased by 22%, and the content distribution increased by 48%. The algorithm in this article improved the recommendation effect.