How Users Perceive and React to Labeled AI-Generated Content on Social Media: A Longitudinal Study of Cognitive and Emotional Effects

As AI-generated content (AIGC) becomes commonplace on social platforms, “AI-generated” labels are increasingly used to calibrate trust and steer sharing. We tested the short-term persistence and emotional side-effects of labeling in a two-wave online experiment (Wave 1 N = 400; Wave 2 N = 256; 18-day interval). Participants viewed health posts varying in veracity (true vs. false) and content type (nonprofit news vs. profit advertising), each carrying the same label. Across waves, perceived accuracy, information credibility, and sharing intention showed no mean change, suggesting attenuated or stable effects within this window. Cross-wave models indicated modest temporal continuity, strongest for sharing intention. Notably, labeling induced a small negative shift in affect after exposure to true content relative to baseline, consistent with an “emotional tax.” Trait self-control buffered this affective cost and showed limited moderation of cognitive outcomes. Findings delineate the near-term bounds of labeling and motivate complementary design strategies.