On September 1st, Douyin experienced a sudden, large-scale malfunction in its recommendation service. A wave of users reported that their home feeds had veered wildly off course, bombarding them with AI-generated short dramas, health tips, parenting advice, and movie recaps—content completely unrelated to their personal tastes. It felt, as many put it, like they’d stumbled into a channel meant for retirees. What made it even more baffling? Even after repeatedly long-pressing videos and tapping “Not Interested,” the same low-quality content kept popping up. Some users even ran into basic functional glitches like empty trending lists and laggy loading screens.
Frustration boiled over quickly, with hashtags like “Douyin Recommendations Are Terrible” and “Douyin Stumbles into Senior Citizen Channel” rocketing up the Weibo hot search list. In response, Douyin’s customer service told Pear Video that they had received the reports, were taking the issue very seriously, and were urgently investigating the root cause. They advised users to try restarting the app in the meantime and asked for patience while they worked on a fix.
Based on feedback from numerous users on social media, this recommendation glitch first appeared on the evening of August 31st and still hadn’t been fully resolved by the following morning. The common thread among affected users? Their carefully cultivated personalized feeds—built on likes, watch history, and interest tags—suddenly seemed to suffer from amnesia. In their place? A flood of homogeneous, low-quality, broadly appealing content.
One young user vented that they usually follow tech, finance, and digital content, but that night, their feed was nothing but health supplements, sentimental quotes for the middle-aged, and AI-generated mini-dramas. “It felt like my dad had gotten ahold of my phone,” they joked. Other users noted the feed seemed to have turned into a random lottery, with almost zero connection to their interests, and some even reported seeing the exact same video pushed to them repeatedly.
The AI short dramas that drew the most complaints are, ironically, one of the fastest-growing content categories on short-video platforms right now. Data shows that domestic AI short drama users surpassed 600 million in the first half of this year alone. On Douyin specifically, native AI-generated dramas and animated series saw over 221,900 new titles launched, amassing a cumulative 515.738 billion views. The problem? A glut of cheaply and hastily made AI content relies heavily on platform algorithms for distribution. So, when the recommendation engine hiccups, this kind of content is quick to flood everyone’s screens.
Back on June 15th, Douyin actually tweaked its AI video traffic distribution rules, setting up a dedicated traffic pool for compliant AI content. Creators who clearly labeled their AI-generated material were rewarded with extra organic reach, while low-quality, mass-produced AI videos using generic templates were throttled. But clearly, the pace of regulation hasn’t kept up with the sheer explosion of supply.
For an app as ubiquitous as Douyin, even a brief, localized hiccup in the recommendation algorithm can seriously impact the experience of tens of millions of users. According to QuestMobile’s “First Half of 2026 China Mobile Internet Power Value Ranking,” Douyin’s main app has hit a staggering 1.029 billion monthly active users, making it only the second super-app in China to cross the billion-user milestone after WeChat. It boasts around 830 million daily active users, who spend an average of nearly two hours per day on the platform, collectively racking up over 20 billion video plays daily.
With such a massive user base, any minor tweak to Douyin’s recommendation algorithm can trigger a cascade of unintended consequences. According to the algorithm principles publicly shared by Douyin’s Safety and Trust Center, the core logic of its recommendation system involves predicting user behavior signals—including completion rate, likes, comments, saves, shares, and even “not interested” feedback. It then pushes the videos with the highest predicted feedback value to the right users. This whole process involves behavioral probability prediction, value model evaluation, and minute-level real-time feedback loops.
At the end of 2025, Douyin completed its largest recommendation system upgrade in recent years, shifting from a single “content” tower model to a dual-tower model that considers both “creator and content.” The system now doesn’t just judge the quality of a single piece of content; it also evaluates the creator’s overall weight, factoring in originality rate, how long fans stick around, interaction depth, content quality history, and fan return visit rate. With such a complex system running daily, any model update, parameter adjustment, or even a server cluster failure could easily lead to some users’ feeds going off the rails.
