Generative AI-enhanced learning experiences for computational thinking: A systematic scoping review and design guidelines

Synthesis: Ouaazki, Shibani, Knight, and Holzer conduct a systematic scoping review of studies (screened from 1,198 initial results) examining how Generative AI is used to support the teaching of computational thinking (CT), and derive design guidelines. The field is young but rapidly growing: most interventions target undergraduate students on basic programming tasks, often using off-the-shelf tools with limited integration. GenAI typically plays one of four roles — coder, tutor, debugger, or ideator — with mixed effects on learning outcomes. A central challenge is the tension between overreliance by beginners, who may offload thinking to GenAI, and under-utilization by advanced learners in complex projects. Seven design guidelines distill how to integrate GenAI for CT effectively while minimizing risks.

Key Findings

• Young but rapidly growing field. From 1,198 articles screened in October 2024, the included studies show most GenAI-for-CT interventions focus on undergraduate students and basic programming tasks, frequently using off-the-shelf tools with limited integration into the course.
• Four recurring GenAI roles. GenAI is typically used as a coder, tutor, debugger, or ideator, with mixed (sometimes null) effects on learning outcomes depending on how the tool is configured and scaffolded.
• The overreliance–under-utilization tension. Beginners tend to over-rely on GenAI, offloading their thinking (see Cognitive Offloading), while advanced learners may under-use it in complex projects — a two-sided design problem that no single configuration solves.
• Coding assistance can support or supplant learning. Using GenAI to generate or debug code can enhance learning when combined with Computational Thinking concepts in open-ended projects, but can undermine skill acquisition when it bypasses the learner's own algorithmic thinking.
• Seven design guidelines (below) provide actionable direction for educators and system designers.

The seven design guidelines

1. Guide GenAI use for beginners — explain the risks and promote tutor-style prompting to support students in using GenAI as-is, while protecting against uncritical acceptance.
2. Confine GenAI interactions for novice and young learners — for teenagers and beginners, use GenAI in a controlled, limited manner.
3. Combine GenAI with CT concepts in open-ended projects — expand GenAI's role beyond solution generation to include coding, debugging, and ideation within tasks that require students to engage CT concepts themselves.
4. Explore advanced learning-experience integration — leverage GenAI in more advanced roles (e.g., as a pair-programmer) to promote higher-order engagement and self-regulated learning (e.g., Socratic-style prompting).
5. Measure outcomes for reflection — embed mechanisms that promote both learner and instructor reflection on GenAI use, since the review finds outcome measurement is often weak.
6. Design learning experiences with digital Ethics and integrity in mind — address privacy, bias, and academic integrity concerns, mitigating issues by design (e.g., privacy-by-design).
7. Rethink targeted skills for CT — evaluate which skills students should actually acquire in CT education, balancing instruction between algorithmic thinking and other CT dimensions.

What this means for practice

• Instructors. For beginners, position GenAI as a tutor that prompts reasoning rather than a coder or debugger that supplies the solution, and explain the over-reliance risk before students touch the tool.
• Instructors. Differentiate Scaffolding by learner level: the same tool needs tutor-style prompting and controlled, limited use for novices and teenagers, and richer open-ended integration for advanced learners, who often under-use it in complex projects.
• Instructional designers. Embed GenAI in open-ended projects that require learners to apply computational thinking concepts themselves, since more than 80% of the reviewed learning experiences used simple exercises and most targeted basic programming in introductory courses.
• Instructional designers. Build reflection into GenAI-supported activities with critical-thinking prompts, post-task debriefings, and learning journals, because only a minority of studies examined how students reflect on their GenAI use or how instructors adapt.
• Researchers. Measure reflection and instructor adaptation alongside outcomes using new scales that separate over-reliance risk from the conditions for success, since the review found outcome measurement often weak.

Limitations

• Scoping review rather than primary evidence: 38 studies from 1,198 records searched in October 2024, so the design supports mapping and guidance, not causal claims.
• Corpus-level methodological weakness: randomized controlled trials remain limited and many findings come from less controlled or exploratory settings, which the authors say may explain the mixed outcomes.
• Model monoculture: the overwhelming majority of reviewed learning experiences used one company's GPT models, limiting comparative insight across models.
• Coverage gaps: the search stopped in October 2024, and records were excluded for being non-English (n = 1) or available only as abstracts (n = 4).

Citation

Ouaazki, A., Shibani, A., Knight, S., & Holzer, A. (2026). Generative AI-enhanced learning experiences for computational thinking: A systematic scoping review and design guidelines. Computers and Education: Artificial Intelligence, 10, 100608.