Thinking Through a Human Cognition Lens : The Effect of Generative AI Processing Time on Consumer Decision- Making Through Effort Inference

다국어 초록 (Multilingual Abstract)

The rapid proliferation of generative artificial intelligence has fundamentally repositioned AI systems from passive information retrieval tools to active participants in consumer decision-making. As consumers increasingly delegate consequential choices to AI advisors across domains ranging from financial planning to healthcare, a critical challenge emerges: they cannot directly assess the quality of AI-generated recommendations because the underlying reasoning process is inherently opaque and inaccessible. We propose that, under such opacity, consumers do not evaluate AI through a technical or computational lens. Instead, they rely on cognitive frameworks developed for human judgment, interpreting AI behavior through a human cognition lens. Within this framework, observable process cues serve as the basis for inference, allowing consumers to estimate underlying reasoning quality. Drawing on the effort–quality heuristic, we focus on AI processing time—the visible delay before a response—as a key procedural cue that signals cognitive effort and shapes evaluation when consumers interpret it as evidence of deliberative reasoning rather than mechanical computation. To test this framework, we conducted five studies employing complementary methodologies to develop and empirically validate a moderated serial mediation model linking AI processing time to consumers' intention to follow AI-generated recommendations. Study 1 provided preliminary naturalistic evidence consistent with the human cognition framing effect through a large-scale analysis of ChatGPT user reviews (N = 84,214), revealing that human-oriented language was approximately 3.38 times more prevalent than machine-oriented language. While this pattern is suggestive of a tendency to frame AI interactions in anthropomorphic terms, the high prevalence of ambiguous keywords (62.99%) and the keyword-based nature of the analysis mean that these findings are best interpreted as preliminary observational grounding rather than definitive evidence of the human cognition lens, thereby motivating the experimental investigations that follow. Study 2 experimentally established the core sequential mediation pathway, demonstrating that when consumers interpret longer AI processing time as reflecting cognitive deliberation, it increases consumers' willingness to adopt AI- generated recommendations mediated by perceived cognitive effort and perceived information quality. Study 3 confirmed the core causal pathway and moreover pinpointed task complexity as a moderating boundary condition, revealing that the sequential mediation pathway is activated exclusively under conditions of high task complexity. Study 4 extended the model by introducing interaction mode as an additional boundary condition, revealing that the sequential mediation pathway emerged exclusively under relational, but not functional, interaction conditions. Study 5 provided the most direct mechanistic test of the proposed framework by manipulating attribution type, demonstrating that the sequential mediation pathway was operative only when AI processing time was attributed to cognitive deliberation rather than mechanical computation, thereby confirming that the human cognition lens constitutes the operative psychological mechanism. Collectively, the findings establish that AI processing time shapes consumer decision-making not through its duration alone, but through the interpretive lens consumers apply to it: when interpreted as evidence of cognitive deliberation through the human cognition lens, longer processing time activates an effort inference process that increases recommendation compliance; when interpreted as mechanical computation, this effect disappears entirely. This study advances theory in three key ways. First, and most fundamentally, it establishes that consumers systematically recruit a human cognition framework when evaluating generative AI, representing a paradigmatic reorientation that departs from prior digital technology research, which consistently treated longer response times as signals of inefficiency. Because generative AI's conversational and linguistically fluent interface structurally resembles human communication, consumers naturally transfer the cognitive schemas they have developed for evaluating human judgment to the AI evaluation context, such that processing time is reinterpreted not as a signal of system inefficiency but as behavioral evidence of deliberative effort—rendering the effort-quality inference heuristic operative in a domain where it was previously untheorized, but only when consumers perceive the AI through a human cognition lens. Third, it delineates three theoretically distinct boundary conditions, namely task complexity, interaction mode, and attribution type, that jointly specify when and why consumers interpret AI processing time through a human cognition lens, converging on the insight that it is not the duration of AI processing time per se, but consumers' interpretation of that duration as evidence of cognitive deliberation, that constitutes the operative mechanism driving recommendation compliance. Together, these contributions offer a conceptually rigorous and empirically substantiated account of the ways in which procedural cues influence consumer reliance on generative AI, with practical implications for the design, deployment, and communication of AI advisory systems across high-stakes consumer domains.

목차 (Table of Contents)

• CHAPTER 1. Introduction 1
• 1.1 Research Background 1
• 1.2 Differentiation from Prior Literature and Research Questions 2
• CHAPTER 2. Theoretical Background 8
• 2.1 Generative AI in Consumer Decision Making 8
• 2.2 Heuristic Cognition of Generative AI 10
• 2.3 Human Cognition Frame in Heuristic Evaluation of Generative AI 20
• CHAPTER 3. Hypothesis Development 22
• 3.1 Processing Time as a Heuristic Cue 22
• 3.2 Boundary Conditions of Effort Inference 26
• 3.2.1 The Moderating Effect of Task Complexity 27
• 3.2.2 The Moderating Effect of Interaction Mode 28
• 3.2.3 Attribution of Processing Time 31
• CHAPTER 4. Overview of Our Studies 37
• 4.1 Study 1 39
• 4.1.1 Data Collection 39
• 4.1.2 Data Preprocessing 40
• 4.1.3 Keyword-Based Category Detection 42
• 4.1.4 Results 44
• 4.1.5 Discussion of Study 1 46
• 4.2 Study 2 49
• 4.2.1 Experiment Design and Participants 49
• 4.2.2 Stimuli and Measures 50
• 4.2.3 Results 52
• 4.2.4 Discussion of Study 2 54
• 4.3 Study 3 56
• 4.3.1 Experiment Design and Participants 56
• 4.3.2 Stimuli and Measures 57
• 4.3.3 Results 60
• 4.3.4 Discussion of Study 3 68
• 4.4 Study 4 70
• 4.4.1 Experiment Design and Participants 70
• 4.4.2 Stimuli and Measures 71
• 4.4.3 Results 74
• 4.4.4 Discussion of Study 4 84
• 4.5 Study 5 85
• 4.5.1 Experiment Design and Participants 85
• 4.5.2 Stimuli and Measures 86
• 4.5.3 Results 88
• 4.5.4 Discussion of Study 5 98
• CHAPTER 5. General Discussion 101
• 5.1 Summary of Findings 101
• 5.2 Theoretical Contributions 105
• 5.3 Managerial Implications 111
• 5.4 Limitations and Future Research 116
• References 126
• Report References 150