AI literacy and graduate employability in the age of artificial intelligence

Synthesis: A survey of 455 final-year undergraduates in Jakarta finds that AI Literacy travels to employment prospects by a specific route: it raises career adaptability, and adaptability in turn is what students read as employability. The direct path is strong on its own, but the mediation result is the paper's argument — technical fluency matters because it strengthens the capacity to adapt, not as a credential in itself. The expected moderator, growth mindset, did nothing: it neither strengthened nor weakened any path.

Key Findings

1. AI literacy predicted perceived employability directly and strongly (β = 0.447, t = 11.295, p < 0.001), and also predicted career adaptability (β = 0.361, t = 7.589, p < 0.001), supporting the first two hypotheses.
2. Career adaptability predicted perceived employability (β = 0.360, t = 7.743, p < 0.001), and it significantly mediated the AI literacy to employability relationship (indirect β = 0.173, t = 5.485, p < 0.001).
3. Growth mindset did not moderate the AI literacy to career adaptability path (β = -0.037, p = 0.202), and the moderated mediation index was also non-significant (0.003, p = 0.908).
4. Effect sizes were medium to large for the main paths — f² = 0.205 for AI literacy on career adaptability, 0.318 on employability, and 0.156 for career adaptability on employability — against 0.004 and 0.000 for the interaction terms.
5. The authors read the null moderation as evidence that AI literacy has become a baseline technical competency for Generation Z, whose effect on career outcomes holds regardless of individual psychological orientation.
6. Data came from 455 final-year students recruited by convenience sampling through lecturers' networks at private universities in Jakarta, analyzed with PLS-SEM; the outcome was perceived employability, a self-rating, not employment measured later.

How the study was built and measured

The sample is 455 final-year undergraduates in the Generation Z cohort, drawn by convenience sampling through lecturers' academic networks at private universities in Jakarta, with an online questionnaire and voluntary participation. The paper reports the split as balanced — 52.10 percent female (n = 237) and 47.90 percent male (n = 218) — with respondents clustering at age 22 (34.95 percent) and 24 or older (26.81 percent), and representing a range of disciplines. Sample size was fixed in advance with a G*Power analysis rather than by what the recruitment happened to yield.

All four constructs — AI literacy, career adaptability, perceived employability and growth mindset — came from previously validated scales adapted for the higher-education AI context, and the questionnaire was deliberately kept brief and mobile-friendly to reduce response burden. The measurement model holds up on the usual criteria: composite reliabilities of 0.911 for AI literacy, 0.900 for growth mindset and 0.898 for career adaptability, with AI literacy's average variance extracted at 0.794, above its correlations with the other constructs.

What the mediation argument claims

The paper's substantive move is to refuse to treat AI literacy as a credential that employers simply reward. Instead it models career adaptability as the mechanism: AI literacy raises students' capacity to adjust goals, plans and self-presentation to a changing labor market, and that adaptive capacity is what converts into perceived employability. The numbers support this ordering, with all three direct paths significant at p < 0.001 and the indirect effect significant at β = 0.173.

The practical reading is that employability work aimed at AI should not stop at tool training. If adaptability is doing part of the work, programs that build it — articulation of transferable skills, rehearsal of career pivots, structured reflection on how a changing task mix affects one's plans — are part of AI readiness rather than a separate careers service function.

The moderator that did nothing

The study's second contribution is a null result. Growth mindset was expected to amplify the AI literacy paths, and it did not: the interaction on career adaptability came out at β = -0.037 with p = 0.202, and the moderated mediation index at 0.003 with p = 0.908. Effect sizes for the interaction terms were 0.004 and 0.000, against 0.205, 0.318 and 0.156 for the main paths — an order of magnitude apart.

The authors explain the mismatch by scale of disruption. When a technology reshapes an entire labor market, believing that ability can be developed may not add much on top of actually being able to adapt. Their own caution is that the null could equally be an artifact — a ceiling effect or restricted variance in mindset scores — and they note that this possibility was not tested. The simple slope analysis supports the reported version: the lines are parallel rather than crossing, so the AI literacy relationship looks the same at every level of growth mindset.

What this means for practice

• Instructors. Teach AI literacy as adaptability rather than as tool proficiency. The mechanism the study supports is students who can revise their goals and plans in response to a changing task mix, so an assignment that has students re-plan their own workflow after a tool changes teaches more than a tool-specific walkthrough.
• Curriculum designers and program leads. If adaptability carries part of the employability effect, build it deliberately: articulation of transferable skills, structured reflection on how AI is changing the tasks in a field, and rehearsal of career pivots belong inside AI literacy provision rather than in a separate careers module.
• Administrators and careers services. Treat a self-rated employability gain as a hypothesis, not an outcome. Students who feel more employable are not yet employed, so pair any AI literacy initiative with an outcome measure — internship supervisor ratings or placement data — before reporting impact.
• Researchers. The null moderation is the most portable finding and the most fragile: it is one convenience sample, and the authors concede a ceiling effect could explain it. Before citing it as evidence that mindset does not matter under technological disruption, test the restricted-variance explanation and replicate across institutions and disciplines.

Limitations

• The data are single-source and self-reported, which the authors name as a common-method-variance, subjectivity and social-desirability risk: students may overestimate their own AI literacy and employability relative to objective industry standards. The employability outcome is a perception, not a job obtained.
• The design is cross-sectional, so it captures perceptions at one point in time and cannot establish causal or temporal direction, even though the model is written as a path from AI literacy through adaptability to employability.
• Sampling was non-probability convenience sampling through lecturers' networks at private universities in Jakarta. The authors state that generalizability to the global Generation Z population is limited by a metropolitan sample with its own socioeconomic pressures and AI adoption rates.
• The non-significant growth-mindset moderation may stem from a ceiling effect or restricted variance in mindset scores, a possibility the authors flag but explicitly did not test.
• No multi-group analysis was run, so variation by gender or academic discipline is untested and the model is reported as one pooled structural estimate.
• The authors note that rapid AI evolution may limit the temporal stability of the findings, since labor-market demands and the AI skills expected of graduates are both still moving.

Citation

Hendrawan, K., & Lestari, S. D. (2026). AI literacy and graduate employability in the age of artificial intelligence: The mediating role of career adaptability and the moderating role of growth mindset. Social Sciences & Humanities Open, 14, 103551.