By BSD –
Curated by Business Science Daily — peer-reviewed sources, human-verified.
Learn more
About Our Curation Process
Business Science Daily curates academic research in business and economics. Each featured study is selected from reputable, peer-reviewed journals, institutional repositories, or working papers (e.g., Elsevier, Sage, NBER, SSRN).
Articles are carefully summarized to ensure clarity and accuracy, with direct citations or links to original sources. Our process emphasizes transparency, academic integrity, and accessibility for a broader audience.
Learn more in our Editorial Standards & AI Policy.
While generative AI tools are widely praised for enhancing creativity, accelerating content creation, and supporting problem-solving, their psychological costs remain largely overlooked. The paper authored by Yang, Zeng, Xing, & Hu, (2026) reveals how the very nature of generative AI interaction creates cognitive and emotional fatigue that undermines both user experience and the systems’ long-term improvement.
The study identifies two distinct forms of interaction-level uncertainty that plague GenAI users: prompt uncertainty (the struggle to phrase effective inputs) and response uncertainty (the unpredictability of outputs even with identical prompts).
These factors are not minor inconveniences but significant psychological burdens that lead to measurable cognitive and emotional fatigue. More critically, this fatigue reduces users’ willingness to provide the feedback that is essential for improving AI systems. This feedback loop creates a self-defeating cycle where the mechanisms for enhancement are undermined by the experience of use.
Therefore, understanding these dynamics is not only about user comfort, but also about ensuring that the promise of generative AI does not falter due to the very human limitations of those expected to use it.
For a detailed exploration of this research, navigate the interactive tabs below.
Fatigued by Uncertainties: Exploring the Cognitive and Emotional Costs of Generative AI Usage
How prompt uncertainty and response uncertainty in generative AI systems lead to distinct forms of user fatigue and reduce feedback critical for AI improvement.
Yang, Zeng, Xing, & Hu, (2026). study examines how two distinct types of uncertainty in generative AI interactions—prompt uncertainty (uncertainty about how to phrase effective prompts) and response uncertainty (uncertainty about how AI will respond even to identical prompts)—lead to cognitive and emotional fatigue, ultimately reducing users’ willingness to provide feedback that’s essential for AI system improvement.
Summary
The paper addresses a critical gap in understanding the psychological costs of human-AI interaction, particularly focusing on how different forms of uncertainty in generative AI systems lead to user fatigue and undermine feedback mechanisms essential for AI improvement.
The Core Problem: While generative AI systems offer immense potential, they introduce unique challenges rooted in their inherent uncertainties. Users face two primary forms of interaction-level uncertainty: (1) Prompt uncertainty – uncertainty about how to phrase effective prompts to get desired responses, and (2) Response uncertainty – uncertainty about how AI will respond even when using identical prompts.
The Psychological Costs: These uncertainties lead to distinct forms of fatigue:
- Prompt uncertainty → Emotional fatigue: Frustration and emotional exhaustion from trial-and-error in prompt formulation
- Response uncertainty → Cognitive fatigue: Mental exhaustion from evaluating unpredictable, inconsistent outputs
The Critical Consequence: Both types of fatigue significantly reduce users’ willingness to provide feedback (rating outputs, reporting errors), which is essential for the iterative refinement of AI systems. This creates a feedback paradox: the very mechanisms needed to improve AI are undermined by the user experience of using AI.
Theoretical Framework: Human-Centered AI Perspective
The study is grounded in the Human-Centered AI (HCAI) framework, which integrates system properties, human values, and human purpose to understand AI interaction dynamics.
1. Two Forms of Interaction Uncertainty
- Prompt Uncertainty: Difficulty in determining how to phrase effective prompts to elicit desired responses. This involves trial-and-error interaction, requiring significant cognitive and emotional resources to navigate input ambiguity.
- Response Uncertainty: Unpredictability of GenAI outputs, where identical prompts can produce responses differing in quality, relevance, or tone. Users must repeatedly evaluate outputs and determine alignment with goals.
2. Dual Dimensions of GenAI-Induced Fatigue
- Cognitive Fatigue: Mental exhaustion from sustained cognitive effort while interacting with GenAI systems, manifesting as difficulty concentrating and diminished analytical capacity.
- Emotional Fatigue: Affective exhaustion characterized by frustration, dissatisfaction, and emotional weariness from managing the emotional demands of uncertain interactions.
3. Research Model & Hypotheses
The study tests six hypotheses linking uncertainties to fatigue dimensions and feedback behavior:
- H1a/b: Prompt uncertainty → Cognitive/Emotional fatigue
- H2a/b: Response uncertainty → Cognitive/Emotional fatigue
- H3a/b: Cognitive/Emotional fatigue → Reduced feedback provision
The model situates these relationships within the HCAI framework: Uncertainty (Properties) → Fatigue (Values) → Feedback Provision (Purpose).
Methodology & Data
Research Design & Data Collection
This study employed a quantitative survey design with data collected from 832 GenAI users in China (September-October 2024):
- Platform: Wenjuanxing (similar to MTurk/Qualtrics), with over 6 million registered users
- Sample Size: 832 valid responses from initial 914 participants after quality controls
- Quality Controls: Screening questions, reverse-coded items, instructional manipulation checks, marker variable approach
Sample Characteristics
- Gender: 46.8% male, 53.2% female
- Age: 36.1% (18-30), 61.3% (31-50), 2.6% (50+)
- Education: 60.6% bachelor’s degree, 19.4% master’s or higher
- GenAI Used: Baidu Ernie (24.6%), TikTok Doubao (20.7%), iFlytek Spark (14.5%), others
- Usage Duration: 35.0% (3-6 months), 31.4% (6-12 months), 21.1% (12+ months)
Measurement & Analysis
Constructs Measured (5-point Likert scales):
- Prompt Uncertainty: 5 items (e.g., “I’m uncertain which way of asking will get high-quality responses”)
- Response Uncertainty: 5 items (e.g., “The GenAI often provides different responses to the same prompt”)
- Cognitive Fatigue: 4 items (e.g., “Using the GenAI leaves me feeling cognitively drained”)
- Emotional Fatigue: 4 items (e.g., “After interacting with the GenAI, I often feel emotionally exhausted”)
- Feedback Behavior: 4 items (e.g., “When I notice errors in the GenAI’s responses, I use the feedback feature”)
Analysis Methods: Confirmatory Factor Analysis, Structural Equation Modeling using Weighted Least Squares Mean and Variance adjusted method for ordinal data.
Key Findings
1. Distinct Pathways from Uncertainty to Fatigue
- Prompt Uncertainty → Emotional Fatigue (β = 0.357, p < 0.05): Supported. Prompt uncertainty primarily induces emotional exhaustion through frustration from trial-and-error input formulation.
- Prompt Uncertainty → Cognitive Fatigue: Not supported (β = 0.043, p = 0.369). Prompt adjustment may involve less sustained cognitive engagement than output evaluation.
- Response Uncertainty → Cognitive Fatigue (β = 0.443, p < 0.05): Supported. Response unpredictability significantly taxes mental resources through repeated evaluation of inconsistent outputs.
- Response Uncertainty → Emotional Fatigue: Not supported (β = -0.082, p = 0.078). Users may normalize output variability as a technical feature rather than personal failing.
2. Fatigue Significantly Reduces Feedback Provision
- Cognitive Fatigue → Feedback Behavior (β = -0.314, p < 0.05): Supported. Mental exhaustion leads users to simplify interactions and neglect secondary tasks like feedback.
- Emotional Fatigue → Feedback Behavior (β = -0.498, p < 0.05): Supported. Emotional drain reduces patience and motivation for additional interaction, even simple feedback actions.
3. Control Variable Insights
- Age: Negatively related to cognitive fatigue (β = -0.084) and feedback provision (β = -0.075) – younger users experience more strain but engage more in feedback.
- Education: Reduced feedback provision (β = -0.077) – more educated users may be less inclined to provide input.
- Usage Duration: Lowered both cognitive (β = -0.083) and emotional fatigue (β = -0.105) while increasing feedback provision (β = 0.108) – experience mitigates strain and sustains collaboration.
4. Model Fit & Reliability
Measurement model showed strong reliability (Cronbach’s alpha: 0.79-0.86) and validity (AVE > 0.50). Structural model fit indices: χ²/df = 3.293, CFI = 0.949, TLI = 0.941, RMSEA = 0.053.
Implications & Future Research
Theoretical Contributions
- Extends HCAI Framework: Demonstrates how system properties (uncertainty) shape human values (well-being) and influence purpose-related behaviors (feedback).
- Dual Pathways of Uncertainty: Distinguishes between prompt and response uncertainty with distinct cognitive/emotional impacts, moving beyond traditional focus on processing uncertainty.
- Novel Fatigue Outcome: Links GenAI-induced fatigue to feedback behavior rather than just dissatisfaction or discontinuance, highlighting critical system-level implications.
Practical Implications for AI Developers
Reducing Prompt Uncertainty:
- Implement real-time input assistance (adaptive prompt templates, autocomplete)
- Provide step-by-step guidance evolving based on user’s task/context
- Develop natural language feedback loops to clarify vague inputs
Addressing Response Uncertainty:
- Display confidence scores or rationale statements alongside outputs
- Introduce interactive refinement features allowing partial output modification
- Enhance output stability through improved model consistency
Encouraging Feedback Despite Fatigue:
- Gamify feedback mechanisms (badges, levels, virtual tokens)
- Offer contextual feedback requests at opportune moments
- Provide personalized benefits (prioritized feature access) for feedback contributions
Limitations & Future Research
- Sample Generalizability: China-only sample limits cross-cultural applicability
- Other Uncertainty Types: Content ownership, data security, and adoption dilemmas not examined
- Broader Outcomes: Other user behaviors (feature exploration, collaborative use) not studied
- Social/Organizational Contexts: Team-based GenAI use and organizational moderating factors not explored
- Longitudinal Dynamics: Cross-sectional design cannot capture fatigue development over time
Future research should test the model in diverse cultural contexts, examine additional uncertainty types and behavioral outcomes, and explore mitigation strategies for GenAI-induced fatigue to ensure sustainable human-AI collaboration.
References
Yang, H., Zeng, Y., Xing, H., & Hu, P. (2026). Fatigued by uncertainties: Exploring the cognitive and emotional costs of generative AI usage. International Journal of Information Management, 87, 103010. https://doi.org/10.1016/j.ijinfomgt.2025.103010
Key Theoretical Frameworks: Human-Centered AI (HCAI), Cognitive Load Theory, Emotion Regulation Theory.