"Oops! ChatGPT is Temporarily Unavailable!": A Diary Study on Knowledge Workers' Experiences of LLM Withdrawal
A 4-day diary study reveals how LLM dependency creates workflow gaps and challenges professional values.
A team of researchers from KAIST and Seoul National University, led by Eunseo Oh, published a study titled "Oops! ChatGPT is Temporarily Unavailable!" The paper, accepted at the ACM CHI EA 2026 conference, investigates the deep dependency knowledge workers have developed on Large Language Models (LLMs) like ChatGPT. The researchers conducted a four-day diary study with 10 frequent LLM users, deliberately observing their reactions and workflow changes during a temporary withdrawal of the AI tools.
The findings revealed that the absence of LLMs significantly disrupted participants' established workflows, exposing critical gaps in their task execution processes. The study conceptualizes LLMs as having become "infrastructural" to modern knowledge work, meaning their use is now an inescapable norm. Interestingly, the forced break led participants to engage in more self-directed work, which allowed them to reconnect with and reclaim core professional values that had been subtly undermined by over-reliance on AI assistance.
- A 4-day diary study with 10 knowledge workers showed ChatGPT withdrawal created significant workflow disruptions.
- The research conceptualizes LLMs like ChatGPT as "infrastructural," meaning their use has become an inescapable norm in professional settings.
- Self-directed work during the withdrawal period helped participants reclaim professional values, suggesting a need for more intentional AI use.
Why It Matters
Highlights the risk of skill erosion and the need for professionals to develop value-driven strategies for AI tool use.