Despite mandatory labeling requirements for artificial intelligence (AI) in marketing and communication, the impact of AI labels on individuals’ cognitive and affective responses remains insufficiently understood. Building on signaling theory, we conceptualize AI labels as informational cues that signal content origin (human versus. AI), thereby shaping users’ inferences under conditions of informational asymmetry. Two experimental studies examine how generative AI labels in social media affect cognitive (i.e. visual attention, credibility) and affective (i.e. psychological reactance, emotions) reactions. An eye-tracking experiment (Study 1, N = 60) using a manipulated Instagram post does not confirm an effect on visual attention, but shows that AI labels increase reactance and reduce credibility. An online experiment (Study 2, N = 220) using the same stimulus replicated these findings and additionally revealed that AI labels spur negative emotions and reduce positive emotions – which spill over to social media engagement. Across both studies, we show that even a small AI disclosure – legally required to increase transparency – can backfire. These results highlight the need for organizations to carefully manage and contextualize AI disclosures to mitigate backlash.
Berner et al. (Sun,) studied this question.