Study: AI Video Fatigue May Just Be Bad Video Fatigue

Short-Video Fatigue Points to Quality Problem, Not AI Alone

A feed packed with videos that look cheap, hollow, or hard to judge can leave a viewer drained. A survey of 513 short-video users in China, published in PLOS ONE, tested three complaints about AI videos: that they look cheap, that they feel repetitive, and that sorting through them takes effort. The cheap-looking worry tracked most closely with fatigue, with the effort of sorting second and the repetitive complaint showing almost no connection.

Researchers Zedai Wang, of Hefei Normal University, and Chao Tang, of Jinling Institute of Technology, wrote that fatigue “is not simply a response to AI-generated content itself.” Viewers may be reacting less to a machine-made label and more to a sense that their feeds are getting worse.

That matters for platforms debating how to handle the flood of AI-generated video in recommendation feeds. If fatigue were a reflex to an AI label, tagging videos might be enough. These results may point to a harder job: improving quality and making feeds easier to sort.

Survey of 513 Adults Linked Low Quality to AI Content Fatigue

China is a useful testing ground, the authors argue, because platforms such as Douyin, Kuaishou, and Bilibili pair feeds tailored by algorithms with fierce competition among creators, which makes them “an advanced global laboratory” for tensions that may show up elsewhere. Wang and Tang posted an anonymous online survey on Chinese social platforms, including WeChat and QQ, for one week in May 2026. Respondents had to be adults in mainland China who watched short videos at least three days a week and had noticed AI-generated clips on their most-used app.

Videos counted whether they were fully machine-made or only heavily AI-assisted, with participants going by platform labels, creator disclosures, and visible clues. Of 536 responses, 23 were thrown out for failing attention checks, questions that tell respondents exactly which answer to pick, leaving 513. Nobody was paid, and responses typically took about ten minutes. Recruitment ran through online networks rather than random selection, so the group does not represent all Chinese viewers. Age, gender, and education were recorded, though the paper’s main text does not break them down.

Beyond asking whether AI content tires people out, the survey split the experience into pieces. Respondents rated how flooded their feeds felt with AI videos, then how strongly they agreed with statements about each of the three complaints, from strongly disagree to strongly agree. Some questions were adapted from earlier published surveys, while others were written specifically for this study. More questions covered overall fatigue, satisfaction, and how likely people were to keep watching similar videos. To separate AI-specific exhaustion from ordinary scrolling burnout, the survey measured general short-video fatigue, attitudes toward AI, and confidence in spotting it.

Perceived Low Quality Had the Strongest Tie to Short-Video Fatigue

Noticing more AI content did not lead straight to exhaustion. That perception was tied to the three complaints, and two of the three carried the link to fatigue. Perceived low quality means awkward visuals, thin storylines, and technical glitches, the kinds of problems the authors say may wear viewers down. Mental overload means the constant work of asking whether a clip is AI, whether it is worth watching, and whether to skip it, so every swipe becomes a small quiz.

General short-video fatigue, the ordinary burnout that comes from scrolling, was tied even more strongly to AI content fatigue than any of the three complaints. Some of the exhaustion may come from scrolling itself.

AI Content Fatigue Came With Lower Satisfaction and Weaker Viewing Interest

People reporting stronger AI content fatigue also reported lower satisfaction and less interest in continuing to watch similar videos. Extra statistical checks followed that pattern back through low quality and overload to the feeling of an AI-heavy feed. The authors expected tired viewers to keep using a platform while skipping AI clips, though the survey measured only interest in watching, not actual behavior.

Study Cannot Prove Low Quality Causes Short-Video Fatigue

A survey captures one moment, so it shows connections, not cause and effect. Viewers who are already tired or dissatisfied may also judge AI videos more harshly, and the study measured perceptions, not actual video quality. Algorithms add another wrinkle: recommendation feeds repeat similar topics and formats whether videos are human-made or AI-made, so some viewers may blame AI for repetition the algorithm created. The weak result for repetitiveness does not prove that formulaic content cannot tire viewers, and the authors note it may reflect overlap with the other measures. Positive reactions such as enjoyment were never asked about, so the survey cannot say whether AI origin plays any role.

Whether labels or other fixes would ease the fatigue remains untested. The authors floated future experiments with more detailed labels, such as fully generated versus AI-assisted, and user controls like “show fewer AI-generated videos” and “prioritize human-created content.” Other proposals include cutting near-duplicate templates and temporarily thinning the share of AI videos a viewer sees. Until someone runs them, this survey points to a quality problem more than an AI problem.

---