AI training and science student teachers’ TPACK in campus-based and distance education: a comparative study

Synthesis: Mnguni et al. (2026) ask whether AI training levels, and the institutional mode of delivery, explain science student teachers' self-reported Technological Pedagogical Content Knowledge (TPACK). A quantitative, non-experimental survey of 186 final-year Bachelor of Education science students at a South African distance education university (n = 97) and a campus-based university (n = 85) found a campus advantage in self-reported TPACK, 64.0% versus 47.4%, but no uniform training effect. AI training predicted self-reported TPACK only at the distance institution, where a five-credit short course was associated with significantly weaker reported TPACK than no training. Across both cohorts pedagogical knowledge was the weakest component and no domain median reached the strongest-agreement Benchmark, so the authors read readiness as co-constructed through institutional affordances rather than training exposure, and argue for context-responsive Teacher Education attentive to the Digital Divide and Equity in the Global South.

Key Findings

1. Campus-based participants reported stronger TPACK than distance participants, 64.0% against 47.4%; lower scores on the 1 to 5 scale index stronger perceived knowledge (1 = strongly agree).
2. Formal AI training was rare in both cohorts: 76.4% of campus-based and 88.7% of distance participants reported none, and mode was not significantly associated with training level (χ² = 7.12, p = .068).
3. AI training predicted self-reported TPACK only at the distance university (F(3, 93) = 3.42, p = .021), where short-course completion predicted weaker reported TPACK than no training (B = 1.83, p = .039).
4. The campus-based model was null: training level did not predict self-reported TPACK (F(3, 85) = 0.615, p = .607), explaining 2.1% of variance.
5. Pedagogical knowledge was the weakest component in both institutions, reported by 53% of campus-based participants, and no component's median reached the strongest possible agreement (p < .001).

Design, Sample and Instrument

A quantitative, non-experimental comparative survey using an educational measurement approach compared natural groups without manipulation. The 186 participants were purposively sampled final-year Bachelor of Education science students preparing to teach Grades 10 to 12 at one distance university (n = 97) and one campus-based university (n = 85), surveyed with a closed-ended questionnaire in 2024. AI training was coded as none; a few lectures under 20 hours; a five-credit short course up to 50 hours; or a full 10 to 12 credit course of 100 to 120 hours. TPACK used seven dimensions rated 1 = strongly agree to 5 = strongly disagree, so lower scores mean stronger perceived readiness; the PK subscale held two items.

AI Training as a Weak, Context-Dependent Predictor

Practitioner takeaway: contact hours are not a readiness guarantee. AI training predicted self-reported TPACK at the distance institution but not the campus-based one, and the only significant coefficient ran the other way, linking short-course completion to weaker self-reported AI-related professional knowledge than no training (B = 1.83, p = .039), while full-course completion trended stronger without reaching significance (B = −1.51, p = .088). The authors suggest brief exposure may raise awareness of AI integration's complexity, prompting more cautious Self-Assessment, while sustained training may supply deeper grounding. They stress this does not show one mode is inherently better and label the analyses exploratory. For curriculum designers, AI policy and teacher preparation, professional development depth matters more than access.

Pedagogical Knowledge as the Persistent Weak Point

Pedagogical knowledge was the least reported domain in both cohorts, at only 53% of campus-based participants, which the authors connect to evidence that teachers overestimate technological abilities while expressing lower confidence pedagogically. Training that foregrounds technical dimensions without pedagogical and content integration risks fragmented professional identities, so professional development should treat Pedagogies and Teaching Strategies as the anchor rather than an add-on. Divergence in TK, PCK and TCK against convergence in PK suggests pedagogical instruction is more standardized across universities while technological and content knowledge vary with tool access and curriculum emphasis, so generic training fits discipline-specific AI work poorly. In science teaching this means coordinating AI with content, learner needs, assessment, Ethics and classroom realities, so Scaffolding and content-specific evaluation sit inside the same skill.

Context-Responsive Training, Equity and the SDGs

The study is framed by SDG 4, on inclusive and equitable quality education and lifelong learning, and SDG 9, on infrastructure and innovation, treating AI-supported science teaching as both pedagogical and infrastructure work. Distance provision is a major route into a historically unequal South African higher education system. Distance education offers flexibility, Accessibility and fewer geographical barriers but depends on resources, digital proficiency and interaction; reduced social presence can weaken participation, while campus provision supports mentorship, Feedback and practical experience. The authors conclude that sustainable AI integration depends on training quality, contextual fit and pedagogical depth, and without infrastructure, ongoing support and equitable access AI risks reproducing existing disparities. They call for larger samples, observed measures and mixed-methods designs.

What this means for practice

• Instructors. Contact hours are not evidence of readiness: short-course completion at the distance university was associated with weaker self-reported Technological Pedagogical Content Knowledge (TPACK) than no training (B = 1.83, p = .039), so pair introductory AI courses with sustained practice.
• Administrators. Fund depth over reach: full-course completion trended toward stronger self-reported TPACK (B = −1.51, p = .088) while brief exposure moved self-assessment the other way.
• Researchers. Use observation or performance measures when asking whether training changed teaching; this study captured confidence only, not classroom AI use.

Limitations

• 186 final-year students from two purposively sampled universities (n = 97 distance, n = 85 campus) produced response rates of 5.3% against 25%, a nonresponse-bias risk the authors flag.
• TPACK was self-reported with no observation of practice, indexing perceived readiness rather than AI integration, and the two-item PK subscale may not capture its breadth.
• Training cells were badly unbalanced (76.4% of campus-based and 88.7% of distance participants reported no AI training), making the regression coefficients exploratory; the cross-sectional design permits no causal claim.

Citation

Mnguni, L., El Islami, R. A. Z., Nuangchalerm, P., Sethole, K., Sari, I. J., Camara, J. S., & Van Bien, N. (2026). AI training and science student teachers’ TPACK in campus-based and distance education: a comparative study. Computers and Education Open, 11, 100410.