Research Article
Digital governance of generative AI in higher education: authentic assessment as a pathway to student engagement and educational equity
Synthesis: Jiang (2026) tests whether institutional AI governance reaches students as anything more than a policy on a website. Two-wave survey data from 644 undergraduate and postgraduate students across research-intensive, teaching-oriented, and vocational-applied universities in eastern, central, and western China were combined with objective coding of each institution's AI policy, and the model was estimated with institution-clustered standard errors. Perceived governance quality predicted authentic assessment — the assessment students actually experience — well beyond mere policy awareness (β = 0.36 against 0.12) and raised explained variance from 7.7% to 19.1%; authentic assessment then fully mediated the path to engagement (indirect 0.150) and partially mediated the path to perceived educational equity (0.076). Digital access remained an independent predictor of equity, which governance could not substitute for.
Key Findings
- Quality of governance, not awareness of it, did the work. Perceived governance quality predicted perceived authentic assessment well beyond policy awareness, and adding it more than doubled the explained variance.
- Authentic assessment fully mediated the engagement path. Governance quality had no significant direct path to engagement once authentic assessment was in the model.
- Equity was only partly mediated. Authentic assessment carried part of the governance–equity association, but a significant direct path remained, so governance shapes equity through more than assessment design.
- Digital access stayed an independent predictor of equity. Its correlation with perceived equity matched authentic assessment's, and it was not significant in the assessment model.
- AI literacy did not moderate the assessment–engagement link. The interaction was non-significant, though conditional slopes rose from low to high AI literacy.
Why policy awareness is the wrong outcome measure
Prior work found that simply knowing an AI policy exists predicts little about what students do, and this study shows why: awareness and governance quality behave as different constructs. Policy awareness correlated with governance quality at only r = 0.31, correlated weakly with engagement, and had a negligible correlation with equity (r = 0.09). Once governance quality entered the model, awareness's coefficient dropped to 0.12 while quality held at 0.36 (cluster SE = 0.048, p < 0.001), and explained variance rose from 7.7% to 19.1%. For an institution asking whether its AI policy is working, the answer is that measuring publication and dissemination tells you almost nothing; what predicts student outcomes is whether students experience the policy as fair, consistent, and connected to how they are actually assessed.
Assessment as the transmission belt
The mediation structure is the paper's substantive contribution. Governance quality had no significant direct path to engagement — its entire association ran through authentic assessment, which fits the assessment literature's claim that students orient their effort toward what counts. If AI governance changes task design, it changes engagement; if it only changes rules, it does not. The partial mediation for equity is equally informative: authentic assessment carried 0.076 of the governance–equity association, but a direct path remained, which means governance also operates through procedural fairness that students perceive independently of the tasks they are set. The author's framing draws on procedural justice and self-determination theory for exactly this reason: fair process is a need, not a courtesy.
Equity has a material floor
The most policy-relevant finding is the negative one. Digital access predicted perceived equity as strongly as authentic assessment did (r = 0.31), and it was not significant in the assessment model, meaning it operates on a separate pathway. No amount of governance quality compensated for it. This separates two agendas that institutions often merge: redesigning assessment so AI use is defensible, and ensuring students have the connectivity and devices to participate at all. A university can score well on the first and still leave equity untouched, which is why the author argues that equity requires attention to procedural and resource conditions rather than to policy alone.
What this means for practice
- Administrators. Judge an AI policy by what students experience in assessment, not by whether they have read it; the awareness coefficient collapsed once governance quality entered the model.
- Assessment designers. Redesign tasks so AI use is visible and defensible within them — authentic assessment carried the entire governance-to-engagement path.
- Policymakers. Pair governance reform with access provision; digital access predicted equity independently of anything the institution's policy did.
Limitations
- Both governance quality and the mediators are student perceptions collected in the same two waves, so causal ordering rests on theory and temporal separation rather than on manipulation.
- Policy maturity was researcher-coded from institutional documents rather than taken from an external register, so it carries the coders' judgment.
- Chinese universities only, across three institution types; governance regimes elsewhere may transmit through different mechanisms.
- Engagement and equity are self-reported, with no behavioral or outcome data linked to the survey.
Citation
Jiang, X. (2026). Digital governance of generative AI in higher education: Authentic assessment as a pathway to student engagement and educational equity. Frontiers in Psychology, 17, 1939243.