"Oops! ChatGPT is Temporarily Unavailable!": A Diary Study on Knowledge Workers' Experiences of LLM Withdrawal

Abstract: LLMs have become deeply embedded in knowledge work, raising concerns about growing dependency and the potential undermining of human skills. To investigate the pervasiveness of LLMs in work practices, we conducted a four-day diary study with frequent LLM users (N=10), observing how knowledge workers responded to a temporary withdrawal of LLMs. Our findings show how LLM withdrawal disrupted participants' workflows by identifying gaps in task execution, how self-directed work led participants to reclaim professional values, and how everyday practices revealed the extent to which LLM use had become inescapably normative. Conceptualizing LLMs as infrastructural to contemporary knowledge work, this research contributes empirical insights into the often invisible role of LLMs and proposes value-driven appropriation as an approach to supporting professional values in the current LLM-pervasive work environment.

• The paper reveals that abrupt LLM withdrawal disrupts workflows, leading to task delays and increased reliance on human assistance.
• The paper employs a four-day diary study and semi-structured interviews to capture immediate, cognitive, and social impacts of LLM unavailability.
• The paper demonstrates that LLM absence forces workers to reclaim ownership of tasks and re-prioritize responsibilities in increasingly AI-dependent roles.

LLM Withdrawal Among Knowledge Workers: Unveiling Infrastructural Dependency

Study Overview and Methodology

This paper examines the entrenchment of LLMs, particularly ChatGPT, in the daily work practices of knowledge workers by exploring the consequences of abrupt LLM unavailability. Leveraging a four-day diary study paired with semi-structured interviews, the researchers systematically investigated the behavioral, cognitive, and socio-professional repercussions of temporary LLM withdrawal among heavy users (N=10) in South Korea. All participants exhibited high instrumental dependency as measured by items from the LLM-D12 scale. The study measured not only immediate disruptions in workflow, but also deeper shifts in values, sense of ownership, and perceptions of inevitability regarding LLM use.

A custom web-based diary platform facilitated real-time data capture during moments of LLM withdrawal:

Figure 1: Web-based diary interface used for capturing real-time reflection and workflow disruption during the LLM withdrawal period.

Post-withdrawal interviews provided further qualitative depth, resulting in a rich, thematically coded corpus analyzed for emergent patterns. This approach aligns with methods utilized in technology deprivation literature and allows for direct parallel to analogous studies in digital detox and social media abstinence.

Empirical Findings

Workflow Gaps and Task Disruptions

Participants described acute disruption stemming from loss of LLM assistance, evidencing that LLMs had transcended their status as optional tools and had become infrastructural. Specific observations include:

• Social Costs of Substitution: Delegating queries to colleagues was perceived as burdensome, reinforcing a tendency to seek assistance from digital agents over human counterparts.
• Search, Reading, and Writing Degradation: Without LLMs, search tasks became labor-intensive, with participants exhibiting lowered standards in reading comprehension and written communication. The friction of synthesizing information from conventional search was notable, as was a willingness to accept lower-quality outputs in the absence of LLM-powered revision.
• Delayed or Abandoned Work: Tasks requiring immediate or complex support were often postponed or abandoned, indicating latent job-task redesign predicated on LLM ubiquity.
• Misestimation of Task Duration: Participants systematically underestimated the extent of LLM acceleration in workflows, experiencing frustration and schedule overruns when forced to proceed without assistance.

These disruptions offer strong evidence for the LLM’s infrastructural integration, supporting the claim that modern informational labor is being tacitly re-architected around instantaneous, context-sensitive AI help.

Value Reclamation and Work Identity

The abrupt removal of LLMs provided a unique affordance for participants to recover lost or obscured values:

• Restored Clarity and Deliberation: Direct engagement with tasks—absent AI mediation—fostered greater clarity in logic construction and decision tracing. Participants were required to engage with underlying task reasoning, revealing misalignments previously masked by LLM-generated outputs.
• Enhanced Sense of Ownership: Outputs produced independently led to stronger personal investment and pride. Participants were able to discern which aspects of their work had become over-reliant on LLMs, prompting strategic reevaluation of their professional skillset and priorities.
• Critical Task Selection: Time and cognitive constraints forced by LLM withdrawal necessitated a reprioritization of which tasks truly warranted personal attention, foregrounding the encroachment of AI into previously core professional activities.

Perceived Inevitability and Social Norms

The withdrawal amplified awareness of both individual and social dimensions of LLM dependency:

• Subjective Incompetence and Learned Helplessness: Many participants explicitly stated they were unable, and in some cases unwilling, to perform tasks autonomously that they had previously always completed with LLM assistance.
• Normalization and Competitive Pressure: LLM use is now considered a baseline job requirement. Not adopting these tools was equated with self-sabotage, and the social environment was described as exerting implicit pressure to embrace LLM-based workflows.
• Disappearance of Task Boundaries: There is no longer a consensus on what types of queries are appropriate for LLM delegation; scope creep was widespread, with even trivial tasks being routed to LLMs.

Theoretical Implications

The results empirically validate the argument that LLMs have undergone infrastructural inversion—with their impact and depth of integration only fully visible in moments of outage or withdrawal. Unlike prior dependency literature which emphasizes individual skill decay, this study frames the dependency through an infrastructural lens, underscoring the collective and systemic nature of adaptation. The normalization of LLM-based assistance reconfigures not only individual capabilities, but also the social contract of knowledge work, leading to altered norms around work quality, efficiency, and agency.

Prospects for Value-Driven Appropriation in LLM Design

The findings indicate that binary framings (use vs. non-use) are no longer tenable for LLM integration. Instead, the authors advocate for value-driven technological appropriation: a deliberate, reflective negotiation of LLM engagement to preserve clarity, ownership, and professional identity alongside efficiency. Drawing from appropriation theory, LLMs must be configured and adopted in ways that are contextually sensitive, allowing workers to retain autonomy over core tasks, rather than defaulting to maximal automation. Precedents in structured prompting and agency-preserving scaffolds from AI-in-education research suggest design strategies that could generalize to broader professional domains.

Limitations and Directions for Future Research

The study's participant cohort was demographically and geographically constrained (South Korea, high LLM use), suggesting the need for replication across heterogeneous populations and contexts to assess broader generalizability. Moreover, the four-day withdrawal window may only capture short-term effects; more prolonged abstinence could surface deeper adaptive responses or shifts in appropriation strategies. Future research should examine long-term strategies for reflective LLM adoption, potentially through interventions that prime users to consider value-aligned engagement and periodic intentional non-use.

Conclusion

Temporary withdrawal from LLMs among frequent users exposes the extent to which these tools now underpin the fabric of knowledge work. The infrastructural role of LLMs, coupled with strong normative pressures for adoption, necessitates a shift from individual skill-decay concerns to questions of social, organizational, and value-based negotiation. HCI and AI system designers must foreground mechanisms for intentional, values-driven appropriation of LLMs, enabling knowledge workers to sustain both agency and efficiency in increasingly AI-mediated professions.

(2603.26099)