Labeling a post as AI-made quietly costs creators their audience
A study of one million TikTok posts by Stephan Carney, Ignacio Riveros and Stephanie M. Tully of the University of Southern California's Marshall School of Business found that AI disclosures reduce user engagement. TikTok has required such labels since September 2023, when it became the first large United States social media company to do so, creating a natural experiment for the researchers.
The content itself scored at similar quality levels with or without AI, and the drop was not driven by distrust of the technology or by the age of the viewer. What eroded was the one-sided attachment users form with the person behind the account. Brands and influencers adopting generative tools book the saved production effort, while the weakened bond, and the reduced willingness to return, sits nowhere in the budget.
Source: Stanford Social Innovation Review · September 22, 2026