When AI says who made it: A meta-analysis of AI disclosure effects on consumer responses
Mincheol Kim · Computers in Human Behavior Reports · 2026
Artificial intelligence (AI) is increasingly embedded in digital communications, yet the behavioral consequences of disclosing AI authorship remain inconsistent across studies. This meta-analysis synthesizes 21 experimental effect sizes (N = 13,582) to estimate the effect of AI disclosure on human behavioral responses—trust, attitudes, and behavioral intentions—and to identify the conditions under which it is strongest. Using a DerSimonian–Laird random-effects model, the pooled effect is significant and negative (Hedges' g = −0.440, 95% CI [−0.532, −0.348]), with complete directional consistency across all 21 effect sizes. Substantial heterogeneity (I 2 = 80.5%) motivated moderator analyses. Task subjectivity emerged as the primary moderator: effects were markedly stronger in subjective contexts such as advertising, service, and chatbot communication (g = −0.522) than in objective informational contexts such as news labeling (g = −0.216; Q_between = 8.27, p = .004), and effects within objective contexts were strikingly homogeneous (I 2 = 0.0%). Sample size was associated with effect magnitude in univariate analysis, but this association was confounded with task context—large-sample studies clustered in objective domains—and did not remain significant when the two were modeled jointly; we therefore interpret it as a reflection of contextual heterogeneity rather than a purely methodological artifact. Grounded in the persuasion knowledge model, domain ownership threat theory, and the machine heuristic, these findings advance a task-contingent model of AI disclosure with implications for transparency design, platform governance, and human–computer interaction research.