The Role of Instructional Guidance in Generative AI-Assisted Learning: Empirical Evidence from Construction Engineering

Synthesis: Investigates how instructional guidance shapes student-AI interaction in construction engineering education. Introduces a five-step prompting framework grounded in Generative Learning Theory (GLT) to guide learner interaction during review activities. Three conditions tested in a controlled experiment: slide-based learning, unprompted AI-supported learning, and prompted AI-supported learning.

Key findings: Performance differences concentrated on tasks requiring explanation and reasoning (higher-order cognitive outcomes). The prompted condition achieved significantly higher open-ended scores (≈2–3 points on 18-point scale, p<0.01). No significant differences in multiple-choice recall across conditions. Unprompted AI use did not outperform traditional slide-based review.

Demonstrates that the effectiveness of AI-supported learning depends critically on how interaction is structured — a simple prompting framework grounded in learning science can significantly improve higher-order cognitive outcomes. Provides a basis for integrating learning science principles into GenAI systems for education.

What this means for practice

• Instructors. Replace open-ended chat with an explicit prompting sequence during review: the five-step framework grounded in Generative Learning Theory (Clarify, Organize, Integrate, Differentiate, and Correct) was the only condition that outperformed both comparators on open-ended work.
• Instructors. Expect the payoff in explanation and reasoning, not recall: the prompted group scored 12.41 on open-ended questions against 9.68 for slide-based and 10.45 for unprompted AI review (F = 7.32, p = 0.0011), while the multiple-choice difference was not statistically significant (F = 2.38, p = 0.099).
• Instructors. Do not count unprompted AI review as a strategy for higher-order tasks — it remained comparable to slide-based learning on open-ended scores even though it used the same system as the prompted group.
• Instructional designers. Build the scaffold into the tool rather than hoping learners supply it: since prompted and unprompted groups interacted with an identical retrieval-augmented system, the measured effect traces to the interaction structure, not to model access.
• Instructional designers. Implement source-grounded verification as the closing step: the framework's final stage has students re-examine their interpretations against cited materials, and the paper attributes the precision gains to that check.

Limitations

• The final sample is 95 participants after 24 of 119 recruited responses were excluded, and the resulting groups are unbalanced (33 slide-based, 29 prompted, 33 unprompted), which the authors note may affect the stability of the statistical comparisons.
• The study sits in a single instructional context — a construction engineering topic delivered online to Michigan Technological University undergraduates with one set of learning materials — which the authors say may limit generalizability to other domains and course formats.
• Learning was measured with a short-term post-intervention test, so longer-term retention and transfer of knowledge were not captured.
• Time expenditure was analyzed descriptively without formal statistical testing, and only a single prompting strategy was compared, leaving efficiency claims and alternative guidance designs unresolved.

Citation

Hou, X., Xiao, B., Liu, H., & Mueller, S. (2026). The Role of Instructional Guidance in Generative AI-Assisted Learning: Empirical Evidence from Construction Engineering Education.