Generative AI as a conditional job resource under job demands in academic knowledge work: directed content analysis using the job demands–resources framework
Citation. Çögenli, M. Z. (2026). Generative AI as a conditional job resource under job demands in academic knowledge work: directed content analysis using the job demands–resources framework. Frontiers in Artificial Intelligence, 9, 1774525. https://doi.org/10.3389/frai.2026.1774525
Summary. Çögenli uses JD-R-guided directed qualitative content analysis of 15 episodes from a de-identified single-user Gemini interaction log generated during routine academic knowledge work. The paper specifies cognitive, structural/strategic, and emotional/psychosocial functions as candidate GenAI job resources, while treating human oversight and data-integrity/traceability routines as boundary conditions. It argues that GenAI can shift rather than reduce demands when verification, correction, and documentation work are substantial.
Key claims
- Across 15 interactional episodes, the analysis mapped GenAI functions to cognitive resources (clarification, explanation, and complexity reduction), structural/strategic resources (sequencing, roadmaps, prioritization, and work organization), and emotional/psychosocial resources (explicit efficacy-reinforcing and action-oriented cues).
- The paper frames GenAI’s resource value as conditional rather than a fixed property of the tool: interaction sequence, work context, human oversight, and data-integrity/traceability routines determine whether its functions become practically usable support.
- In episodes requiring substantial verification or traceability work, GenAI could shift demands to correction, documentation, and accountability work rather than reduce them.
- The study separates observable AI-output functions from inferred user experience and does not test whether the proposed resources improve engagement, strain, health, or performance.
Evidence & limitations
- This is a peer-reviewed, open-access single-case qualitative analysis of one author’s de-identified interaction log; it specifies candidate mechanisms and boundary conditions rather than estimating prevalence, causal effects, or population-level outcomes.
- The author was both researcher and data source, and the analytic corpus was 15 selected functionally relevant episodes from academic work with Google Gemini; user expertise, prompting style, and context may shape the observed functions.
- The raw transcript contains sensitive personal and health-related information and is not public; the public article and de-identified episode-level supporting materials are the basis for this record.
- Public provenance is verified: the Drive PDF exactly matches the Frontiers public PDF by SHA-256 (2121eda8e4d10405bbd2948bdae7848831162559e3fe7304b8cd075b7c7a0f63).