Does AI at Work Increase Stress? Text Mining Social Media About Human–AI Team Processes and AI Control
Citation. Klonek, F., & Parker, S. (2025). Does AI at work increase stress? Text mining social media about human–AI team processes and AI control. Journal of Organizational Behavior. https://doi.org/10.1002/job.70000
Summary. Klonek and Parker analyze 2,704 ChatGPT-related tweets mentioning teamwork to test whether AI control and human–AI team processes have distinct associations with stress. Using LLM-assisted text coding validated against human ratings, they find that higher AI control is associated with higher stress, while transition and action team processes are associated with lower stress. Action and interpersonal processes also buffer the positive association between AI control and stress, suggesting that AI-mediated teamwork can act as a work-design resource even when AI control is elevated.
Key claims
- AI control should be separated from human–AI team processes rather than treated as the opposite pole of a single automation–augmentation continuum.
- In the tweet sample, AI control was positively associated with stress (β = 0.24, p < 0.001 in the main-effects model; β = 0.22, p < 0.001 in the interaction model).
- Human–AI transition processes (β = −0.13, p < 0.001) and action processes (β = −0.16, p < 0.001) were associated with lower stress, while interpersonal processes did not have a significant main effect.
- Action processes and interpersonal processes moderated the AI-control/stress relationship: the AI-control slope was smaller when action processes were present (B = 0.32 vs. 0.53 when absent) and when interpersonal processes were present (B = 0.36 vs. 0.49 when absent).
- Human–AI team processes were common in the sample: action processes appeared in 69% of analyzed tweets, transition processes in 56%, and interpersonal processes in 23%.
Evidence & limitations
- Evidence comes from a peer-reviewed open-access Journal of Organizational Behavior article using 2,704 tweets from January–March 2023 that mentioned ChatGPT and teamwork-related terms.
- The study uses LLM-assisted text mining with human validation to infer AI control, team processes, and stress from social media text, so the measures are indirect and bounded by what users chose to post publicly.
- The design is tweet-level and cross-sectional; it supports associations among AI control, team-process language, and stress language, not causal claims about AI systems increasing or reducing worker stress.
- The teamwork subsample is useful for human–AI process analysis but not representative of all workplace AI use, all workers, or durable organizational implementations.