Source Transparency in Human–AI Communication: Context-Dependent Effects in Emotional Support and Problem Solving

Abstract

This research examined how source disclosure and task context shape evaluations of generative AI. Across two within-subject experiments, participants evaluated GPT-4o and human responses before and after source disclosure in emotional-support and analytic problem-solving settings. In Study 1, participants evaluated replies to their own positive, neutral, and negative experiences. When sources were hidden, human and AI replies were rated similarly; after disclosure, human responses were evaluated more favorably, whereas AI responses were discounted. Emotional valence moderated some judgments. In Study 2, participants evaluated explanations for logical reasoning problems and re-entered their answers. Disclosure selectively increased evaluations of human explanations but left AI evaluations largely unchanged. Accuracy improved substantially after exposure to correct explanations, regardless of source. Together, the findings show that source transparency shapes subjective evaluations in a context-dependent manner, while answer revision in analytic tasks depends primarily on explanatory content rather than source identity.