Research Article
Examining the relationships between artificial intelligence literacy, AI-TPACK, job satisfaction, and well-being among teachers through structural equation modeling
Synthesis: Kayalı, Balat and Yavuz surveyed 350 teachers in Turkey and modeled how AI literacy, AI-TPACK, job satisfaction and well-being connect. Literacy fed every competence component, but the components then split: knowing how AI changes a subject and how to teach it raised satisfaction, while technical knowledge of AI on its own lowered it, and satisfaction carried whatever effect reached well-being. AI professional development earns its keep when it changes how a task is done, not when it adds another tool to keep up with.
Key Findings
- AI literacy is the upstream resource — it predicts AI-specific competence strongly and traditional subject and pedagogical knowledge more weakly, making it a base rather than a substitute for professional knowledge.
- The component matters more than the total — subject-specific AI knowledge, the kind that connects a tool to a discipline and a lesson, is associated with higher job satisfaction.
- Technical knowledge alone goes the other way — knowing what AI systems are and can do, without a pedagogical use for it, is associated with lower satisfaction.
- Generic pedagogical knowledge also helps — of the traditional domains, only pedagogical knowledge carried satisfaction; subject knowledge and pedagogical content knowledge did not.
- Satisfaction is the only route to well-being — with satisfaction in the model, competence domains have no direct effect on emotional, psychological or social well-being.
- The model explains little of satisfaction — most of what makes teachers satisfied with their work lies outside the competence chain, so the study identifies a mechanism rather than the main cause.
Two kinds of technological knowledge
The interesting result is not that AI literacy matters. It is that two faces of technological competence point in opposite directions. AI-TK, the technical knowledge of what AI systems are and can do, is associated with lower job satisfaction. AI-TCK, understanding how AI changes the content and practice of a subject and which tool fits which purpose, is associated with higher satisfaction.
The coefficients are worth reporting precisely. Literacy predicted AI-TCK at β = 0.65, AI-TK at 0.66, AI-TPCK at 0.73 and AI-TPK at 0.71, all p < .001, against weaker paths to pedagogical content knowledge at 0.32, pedagogical knowledge at 0.35 and subject knowledge at 0.40. On satisfaction, AI-TCK was positive at β = 0.41 and AI-TK negative at β = −0.30, with pedagogical knowledge positive at 0.22. AI-TPCK and AI-TPK were not significant.
The authors read the split through job demands–resources theory: a competence that makes professional life more manageable works as a resource, while one that adds an expectation without adding control behaves as a demand. Knowing more about the technology, in that reading, can mean knowing more about what you are not keeping up with.
Why integrated knowledge did not pay off
The null paths are as informative as the significant ones. AI-TPACK's highest-order component, AI-TPCK, and AI-TPK did not predict satisfaction, and neither did pedagogical content knowledge or subject knowledge. The authors explain this through enactment: teachers may hold integrative knowledge yet be unable to convert it into a rewarding professional outcome because of infrastructure, time or restricted opportunity to implement.
That qualifies the assumption that more AI-TPACK competence will produce better professional outcomes. A competence confers resource value only when it can be used in daily practice; knowledge that cannot be deployed neither adds resources nor imposes demands, so its net association approaches zero. The implication is uncomfortable for programs that count completion: the same literacy that shows up as progress on a log can coexist with no change in how the job feels.
Satisfaction is the gate, and a narrow one
Competence did not predict well-being directly. Satisfaction predicted emotional well-being at β = 0.42, psychological at 0.68 and social at 0.57, and the competence domains reached well-being only through it: indirect-only mediation. AI-TCK's indirect effects ran from 0.18 to 0.28 across the three dimensions; AI-TK's ran from −0.13 to −0.20.
The honesty is in the variance. The model explains 11 percent of job satisfaction, and out-of-sample predictive relevance for well-being is at or near zero, even though the paths from satisfaction to well-being are substantial. The authors say plainly that most of what determines teacher well-being lies outside their model — pay, workload, class size, leadership, subject status. Satisfaction is a mechanism the study can see, not the main cause it can claim.
What this means for practice
- Instructors: Judge a new AI competence by whether it reduces work on a real task; a tool that does not change how you plan, teach or assess adds a demand without a resource.
- Faculty developers: Build AI professional development around subject-specific lesson design and collaborative inquiry rather than tool tours; the survey-course model is the one associated with lower satisfaction.
- Administrators: Measure AI integration by changes in workload and role clarity as well as adoption; training completion is not the outcome, and satisfaction with the work is the pathway the model finds.
- Researchers: Test the direction with a longitudinal design or an intervention that raises AI-TCK specifically, since a cross-section cannot separate the resource account from the selection account.
Limitations
- Cross-sectional survey of 350 in-service teachers in Turkey analyzed with PLS-SEM, so the directional chain is theorized rather than observed.
- All constructs are self-reported at one time point, leaving common-method variance and reverse causation open.
- The model explains only 11 percent of job satisfaction, and predictive relevance for well-being is near zero, so the mechanism it identifies is small against unmeasured workplace factors.
- The AI-TPACK instrument measures reported competence, not observed practice, and results are not broken down by school type, so level-specific patterns are untested.
Citation
Kayalı, B., Balat, Ş., & Yavuz, M. (2026). Examining the relationships between artificial intelligence literacy, AI-TPACK, job satisfaction, and well-being among teachers through structural equation modeling. Acta Psychologica, 268, 107297.