“Made With AI”: The Impact Of Source Labeling On User Perception Of Social Media Ads

Abstract

As generative artificial intelligence (AI) becomes increasingly integrated into digital advertising practices,
understanding how audiences evaluate AI-generated content compared to human-created designs has gained both theoretical
and strategic relevance. This study examines user perceptions of social media post designs produced either by generative AI
technologies or by human designers, and investigates how the presence of a “Made with AI” label influences these evaluations.
Using a 2 (content source: AI-generated vs. human-generated) × 2 (labeling: labeled vs. unlabeled) experimental design,
undergraduate students from five universities in Turkey (N = 388) were exposed to systematically manipulated social media
advertisements. Two posts were created exclusively using generative AI tools through prompt-based production, while two
were produced by professional designers without AI assistance. Labeling conditions were experimentally varied to distinguish
between actual and presented production sources. Participants evaluated the designs across six dimensions: trustworthiness,
purchase intention, creativity, visual attractiveness, professionalism, and perceived quality, and also reported their general
attitudes toward AI technologies. The results indicate distinct evaluative patterns. AI-generated designs were rated more
positively on functional and technical dimensions, including trustworthiness, professionalism, purchase intention, and
perceived quality, whereas human-generated designs received higher evaluations in creativity and visual attractiveness. The
effect of AI labeling was limited and dimension-specific, influencing perceived quality but not broader affective or aesthetic
judgments, and varied depending on the content source. Attitudes toward AI showed consistently positive main effects across
all evaluative dimensions but did not significantly moderate the effects of content source or labeling. Overall, the findings
suggest that users apply different evaluative logics to algorithmic and human design production, while AI attitudes function as
a generalized evaluative orientation rather than a source-specific filter.