Defining AI Fatigue in Academic Contexts: Dimensions, Indicators, and a Stage-Based Model Using Grounded Theory

Technical Summary: Defining AI Fatigue in Academic Contexts

Problem Statement
The rapid integration of Artificial Intelligence (AI) tools into academic settings has introduced a distinct form of strain that existing frameworks—such as technostress, digital fatigue, and Zoom fatigue—fail to fully address. While prior research establishes that digital environments generate overload and mental strain, these models do not account for the specific interaction cycle inherent to AI-supported work: the repeated prompting, reviewing, and verification of rapid, lengthy, and often inconsistent AI-generated outputs. Existing instruments measure general digital strain or academic fatigue but lack the capacity to evaluate the specific cognitive, motivational, emotional, and attentional pressures arising from sustained reliance on generative AI. Consequently, there is no coherent conceptual model explaining how these pressures accumulate or a distinct construct defining "AI fatigue" in higher education.

Methodology
This study employed a qualitative research design guided by Grounded Theory to develop a conceptual model directly from participant accounts.

• Participants: The sample consisted of 1,054 university students (507 male, 547 female; mean age 19.21) from three universities in Pampanga, Philippines, who had used AI chatbots for academic activities.
• Data Collection: Data was gathered through three channels: face-to-face interviews (15–20 minutes), virtual interviews, and online surveys via Google Forms. Participants responded to six open-ended questions exploring their experiences with AI tools, the specific tasks performed, feelings of fatigue or strain, motivational changes, and shifts in learning habits. Responses were collected in both English and Filipino.
• Data Analysis: Researchers utilized the three-stage coding procedure described by Corbin and Strauss (2008):
  1. Open Coding: Independent researchers assigned descriptive labels to phrases reflecting strain, difficulty, or discomfort.
  2. Axial Coding: Labels were grouped into clusters representing recurring forms of pressure (indicators).
  3. Selective Coding: Broader categories were identified to connect indicators into coherent dimensions. A third researcher resolved discrepancies between the two primary coders.
  4. Model Development: The sequence in which participants described these pressures informed a stage-based framework.

Key Contributions and Results
The analysis yielded two primary contributions: a set of five defining dimensions of AI fatigue and a stage-based conceptual model.

1. Five Dimensions of AI Fatigue
The study identified five distinct dimensions, each supported by two specific indicators grounded in participant responses:

• Cognitive Overload (310 responses): The most frequently reported dimension.
  • Output Overwhelm: Mental strain from processing the volume and density of AI responses.
  • Verification Strain: Exhaustion from repeatedly checking AI outputs for accuracy and consistency due to hallucinations or inconsistencies.
• Motivational Disengagement (270 responses):
  • Initiative Loss: A tendency to bypass independent thinking and rely immediately on AI assistance.
  • Competence Doubt: Erosion of academic confidence and self-efficacy due to over-reliance.
• Moral Unease (190 responses):
  • Authorship Guilt: Discomfort and guilt regarding the submission of AI-generated content as one's own work.
  • Understanding Gap: Distress arising from the disparity between the quality of AI output and the student's actual level of understanding.
• Physical Strain (120 responses):
  • Visual Fatigue: Ocular and cranial discomfort from extended reading of dense AI text.
  • Session Exhaustion: Physical depletion following prolonged AI interaction sessions.
• Attentional Drift (125 responses):
  • Pace Impatience: Reduced tolerance for the slower pace of traditional learning materials (e.g., textbooks) after exposure to rapid AI responses.
  • Restless Switching: Fragmented attention and habitual tab-switching driven by a preference for faster answers.

2. The AI Fatigue Model
The study proposes a six-stage, stage-based framework explaining how these pressures develop and reinforce one another across repeated AI interactions:

1. Initial Effort Reduction: AI lowers early cognitive load, becoming a primary response to academic demands.
2. Expansion of AI Roles: Use expands to complex tasks, exposing students to longer, denser outputs.
3. Accumulation of Strain: Pressures across dimensions (particularly cognitive and physical) begin to accumulate.
4. Disruption of Self-Regulation: Students struggle to maintain study routines and resist reliance, experiencing initiative loss and confidence doubt.
5. Fatigue Consolidation: Strain becomes persistent; moral unease and attentional drift become prominent alongside cognitive and motivational pressures.
6. Feedback Loop: Continued AI use returns the student to earlier stages, reinforcing and intensifying all pressures in a cyclical manner.

Significance and Claims
The paper claims to establish the first conceptual and exploratory foundation for AI fatigue as a distinct construct, differentiating it from general technostress or digital fatigue. The authors assert that unlike existing models which treat technology strain as a general response to demands, the AI Fatigue Model specifically addresses the unique interaction cycle of prompting, verifying, and cycling through AI-generated content.

The study positions its findings as a basis for:

• Theoretical Development: Providing a framework for future instrument validation and scale development.
• Empirical Inquiry: Enabling cross-contextual research, particularly in regions like ASEAN where AI adoption outpaces institutional readiness.
• Instructional Application: Guiding educators to identify points in the academic task cycle where pressure accumulates and to design interventions (e.g., structured reflection) that address psychological and motivational pressures.

The authors maintain a modest stance, acknowledging that the findings are exploratory, limited to a specific regional sample, and based on self-reported data. They explicitly state that the model requires confirmatory factor analysis and validation across varied cultural and institutional contexts before broader application.