At Meta's Q2 2026 earnings call, it was disclosed that Instagram's user session length saw double-digit growth year-over-year, driven by an upgraded AI recommendation system. CEO Mark Zuckerberg and CFO Susan Li stated that generative AI has become a core technology for improving content distribution efficiency and user engagement, and the company will further expand this recommendation framework to Facebook's main feed in the future.
According to the report, Meta has used large language models (LLMs) to perform semantic analysis on all public posts in Reels and Feed, identifying content themes, semantics, and potential emotions. Combined with the Muse model, which focuses on video classification and summarization, the system gains a deeper understanding of content. The system then generates more accurate personalized recommendations in real time, relying not only on traditional behavioral signals like clicks and likes but also on users' historical interaction data.
The largest Reels sorting algorithm upgrade so far has already yielded significant results, increasing total session volume by 15 basis points and significantly boosting video sharing rates and viewing duration. The company believes that more accurate AI recommendations can not only enhance user experience but also help creators reach a broader audience and increase ad display opportunities.
However, AI-driven recommendation strategies continue to face regulatory pressure. Meta is currently facing lawsuits from over 20 states in the U.S., alleging that its platform design may lead to youth addiction. The company disclosed that legal expenses related to these lawsuits have reached $2.4 billion, and several cases involving youth protection are expected to go to trial later this year. As generative AI becomes deeply integrated into social media recommendation systems, how to balance enhanced personalization, business growth, and user well-being has become a key focus of ongoing industry attention.
