Designing Teaching Materials with GenAI: Perceptions of Cognitive Load and Learner Competencies in Higher Education

Synthesis: Navas-Romero, Zakharova-García, Guillén-Gámez and Cebrián-Cifuentes surveyed 451 students in education programs at one public university in southern Spain about what they believe generative AI demands of them as designers of teaching materials. Framed by Cognitive Load Theory and TPACK, the 31-item instrument separated perceived usefulness from three learner-related dimensions (metacognitive skills, prior knowledge, problem-solving skills) and three load dimensions (intrinsic, extraneous, and germane). Participants endorsed the practices and the tool, with female students scoring significantly higher on the three learner-related dimensions, though effect sizes were small (r = 0.119 to 0.147) and women made up 82% of the sample. No gender difference reached significance on any load dimension, and the only age or admission-grade association that held was a weak negative correlation between age and extraneous load (r = -0.097). The study measures beliefs about practices, not demonstrated competence, and the authors are explicit about that boundary.

Key Findings

1. Students endorsed GenAI as a design resource. Both groups agreed with statements about the usefulness of GenAI for planning, design and evaluation tasks. The highest perceived usefulness item for female students was that a well-functioning tool would reduce the time spent designing resources and free attention for instructional design (M = 5.20, SD = 1.67); for male students the highest was that GenAI would increase the effectiveness of their future teaching performance (M = 5.05, SD = 1.66).
2. Verification scored highest on problem solving. The top item in that dimension for both genders was the belief that incorrect information generated by GenAI should be checked against other sources before being incorporated into teaching content: female students M = 5.88, SD = 1.52, male students M = 5.22, SD = 1.71.
3. Critical review topped metacognition. The highest metacognitive score among female students was the belief that GenAI output should be critically reviewed so it does not reproduce conceptual errors or biases (M = 5.82, SD = 1.65).
4. Small gender gaps on three dimensions. Female students scored higher on metacognitive skills (5.66 versus 5.20, U = 11,708.500, Z = -3.117, p = 0.002, r = 0.147), prior knowledge (5.61 versus 5.17, U = 11,929.500, Z = -2.890, p = 0.004, r = 0.136) and problem-solving skills (5.45 versus 5.08, U = 12,294.000, Z = -2.537, p = 0.011, r = 0.119). Perceived usefulness did not differ significantly (5.08 versus 4.85, p = 0.134).
5. No gender difference in cognitive load. Intrinsic load (female 3.72 versus male 3.86, p = 0.363), extraneous load (4.70 versus 4.41, p = 0.095) and germane load (5.01 versus 4.88, p = 0.415) all failed to reach significance, and the global CLT score was nearly identical (p = 0.550).
6. Revision demand clustered in extraneous load. The strongest extraneous-load items were that generated text usually needs many corrections before it can be used for educational purposes (female M = 4.84, male M = 4.47) and that generated images usually need many revisions (4.89 and 4.70).
7. Age and admission grade were nearly inert. Only age with extraneous load was significant (r = -0.097, p = 0.040, R² = 0.009). Among female students, admission grade correlated with problem-solving skills (r = 0.133, p = 0.011), metacognitive skills (r = 0.156, p = 0.003) and prior knowledge (r = 0.154, p = 0.003); among male students only metacognitive skills (r = 0.249, p = 0.027).

Study Design

The study was quantitative, non-experimental, ex post facto and cross-sectional. Variables were observed without manipulation, with gender, age and admission grade treated as pre-existing characteristics, and data collected at a single point during the 2025/2026 academic year.

Participants

A non-probabilistic purposive sample recruited 451 students on education-related bachelor's and master's programs at one public university in southern Spain. Women made up 82% (n = 370) and men 18% (n = 81). Mean age was 20.98 years (SD = 4.51) for women and 21.58 years (SD = 5.88) for men. By program, 36.4% were in Early Childhood Education (n = 164), 32.4% in Primary Education (n = 146), 16.6% in Pedagogy (n = 75), 8.4% in Social Education (n = 38) and 6.2% in a master's program in Secondary Education Teaching, Vocational Training and Language Teaching (n = 28). Self-reported prior experience with AI tools was none for 8.9% (n = 40), basic for 79.8% (n = 360) and advanced for 11.3% (n = 51). Frequency of use, specific tools and formal training were not collected.

Instrument

The researchers used an existing instrument, developed and validated in a prior study by Guillén-Gámez and colleagues, rather than building a new one. It carries 31 items across seven dimensions on a seven-point Likert scale from strongly disagree to strongly agree: metacognitive skills, prior knowledge, problem-solving skills, perceived usefulness, intrinsic cognitive load, extraneous cognitive load and germane cognitive load. Cronbach's alpha in the original validation ranged from 0.815 to 0.949 across dimensions, with average variance extracted above 0.50 in every dimension, and those values are reported here as documentation rather than re-estimated. Crucially, the items ask students what they believe is necessary when designing materials with GenAI, so scores are endorsements of beliefs, not evidence of demonstrated skill, observed practice or measured cognitive effort.

Analysis

Because the normality assumption was not met, group comparisons used the Mann-Whitney U test, with effect sizes calculated as r = |Z|/N against N = 451 and interpreted against 0.10, 0.30 and 0.50 as small, moderate and large. Correlations used Pearson's r, justified by the continuous variables, the seven-point scale and the sample size, with R² reported for significant associations. Load dimensions were correlated with age and admission grade in the full sample; perceived usefulness and the three learner-related dimensions were correlated separately by gender, and those gender-specific analyses were exploratory and never formally compared.

What this means for practice

• Instructors. Teach reviewing AI-generated materials as the core task, not as a preliminary, since students in this sample scored the need to verify generated content against other sources and to critically review it for conceptual errors and biases above every other practice.
• Instructors. Give students an explicit revision prompt that asks them to justify each change they make to AI-generated material, because the authors propose justification as the move that turns revision into reflection on pedagogical decision-making.
• Instructors. Plan for the load that comes after generation: the highest extraneous-load items were corrections to generated text and revisions to generated images, and the germane-load items students endorsed most strongly concerned analyzing the limits and responsibilities of using AI in education.
• Faculty developers. Build training that pairs GenAI use with evaluation and adaptation exercises, comparing AI-generated materials with reliable sources and checking them against specific learning objectives, since the study reports beliefs and says the effectiveness of such activities was not tested.
• Researchers. Test whether stated beliefs hold up in observed material-design tasks, since the instrument measures what students think they should do, not the accuracy or pedagogical suitability of the revisions they actually make.

Limitations

• The design is cross-sectional and entirely self-reported, so it establishes no causal relationships and demonstrates neither competence nor cognitive effort during a task.
• The sample was 82% female (370 of 451) and drawn by non-probabilistic purposive sampling from a single university in southern Spain, which the authors name as the main limit on generalizability.
• Several significant results had small effect sizes, and the gender-specific correlations were exploratory and never formally compared, so differences in statistical significance between groups cannot be read as differences in the strength of association.
• The predominantly female sample mixed students from several degree programs and educational levels, including relatively small subgroups, so the results cannot be assumed to represent each program or to establish differences between them.
• Some items refer specifically to future teaching roles, which the authors note may fit students in Pedagogy and Social Education less well.
• Previous AI experience was captured only as none, basic or advanced, with no data on frequency of use, specific tools or formal training.

Citation

Navas-Romero, C., Zakharova-García, G., Guillén-Gámez, F. D., & Cebrián-Cifuentes, S. (2026). Designing teaching materials with GenAI: Perceptions of cognitive load and learner competencies in higher education. Education Sciences, 16(10), 1736.