GenAI Shows Promise as a Teaching Scaffold, but Direct Evidence for Computational Thinking Instruction Remains Thin

A systematic review of how generative artificial intelligence supports instructional design has found that the technology is widely used for lesson planning, idea generation, and feedback, but that direct evidence it improves the teaching of computational thinking is still preliminary. Among 26 studies analysed, only two explicitly embedded computational thinking in the instructional design task, while fifteen offered implicit alignment and nine had no identifiable computational thinking integration at all.

The review, conducted using PRISMA procedures and synthesising literature published between 2023 and 2026, examined how pre-service teachers and related participant groups use generative AI tools such as ChatGPT in lesson design. The authors searched Scopus and ScienceDirect with an open-access criterion, screened records against empirical eligibility, and extracted data on the AI tool used, the instructional design focus, participant experiences, and key findings. Only two studies met the strict criterion that computational thinking be explicitly named and embedded in the design task, intervention, or reported analysis.

This uneven evidence base shapes the review’s central conclusion. The authors treat the fifteen implicitly aligned studies as contextual evidence rather than direct proof of computational thinking instructional design, and the nine studies without identifiable integration as broader background on generative AI in lesson planning. Stronger claims can be made about how generative AI supports general lesson planning; claims about computational thinking-specific design remain anchored in the small explicit subset.

Within that subset, generative AI appears useful as a scaffold between abstract concepts and classroom practice. Pre-service teachers used it to generate initial ideas for computational thinking integration, refine those ideas for real classroom conditions, and clarify conceptual links that might otherwise be equated narrowly with programming. Structured prompting frameworks such as CRAFT and the 6E Learning by Design model further operationalised this support by staging prompt construction, reflection, and design activity.

Across the broader literature, generative AI was most often positioned as a conditional pedagogical scaffold grounded in TPACK and Intelligent-TPACK perspectives. These frameworks emphasise that effective technology use depends on integration with pedagogy and content rather than on the tool alone. Applied to generative AI, the scaffold is conditional: the system may supply ideas, structure, examples, or feedback, while the teacher supplies goals, subject knowledge, contextual judgment, ethical evaluation, and final pedagogical decisions.

Several studies reported gains or positive experiences related to efficiency, confidence, engagement, creativity, or instructional design competence. The review cautions that these constructs were measured in heterogeneous ways across primary studies and should not be generalised as improvements in computational thinking teaching ability. Increased interaction, more prompts, or greater confidence are not treated as direct evidence of improved cognitive ability unless a study used an outcome measure designed to assess that construct.

Time savings emerged as one of the most frequently reported affordances, but the evidence reveals an efficiency paradox. Time saved during initial generation can be offset by the time required to verify information, correct errors, contextualise generic output, or refine prompts repeatedly. Efficiency therefore depends on both the quality of the AI output and the user’s capacity to diagnose what needs revision, reinforcing the need to evaluate it as part of a complete design process rather than by the speed of initial content generation.

Accuracy and ethics concerns recur throughout the included studies. Participants reported inaccurate, generic, inconsistent, or contextually inappropriate outputs, and in the computational thinking-explicit evidence, AI-generated responses sometimes blurred or misrepresented concepts, reinforcing the need for conceptual verification before instructional use. Ethical issues included academic integrity, plagiarism, professional responsibility, bias, and insufficient responsiveness to classroom context. Studies also documented risks of uncritical acceptance, overreliance, and what researchers describe as metacognitive laziness, a pattern especially consequential in teacher education because instructional design is itself a professional learning task.

More productive use of generative AI was consistently linked to human foundations: AI literacy, pedagogical and subject-matter knowledge, structured frameworks, iterative prompting, and critical evaluation. Users with stronger knowledge engaged more critically with outputs, while those with weaker foundations were more likely to accept content at face value. These associations do not establish that iterative prompting causes higher cognitive ability; rather, they show that existing competence shapes how effectively a learner uses the technology and detects limitations in generated content.

The review acknowledges important limitations. The search was restricted to Scopus and ScienceDirect with an open-access criterion, potentially omitting studies indexed in ERIC, Web of Science, ACM Digital Library, IEEE Xplore, and other sources, and creating accessibility bias. Reproducibility was constrained by gaps in the retained review record, including exact search dates, per-query counts, and study-level appraisal scores. The authors conclude that generative AI is better positioned as a provisional pedagogical scaffold than as a reliable autonomous partner, and they call for classroom- and practicum-based studies that explicitly define computational thinking integration, use validated assessment tools, and examine whether observed benefits persist beyond immediate AI-assisted tasks.

Subject of Research: Systematic review of generative AI in instructional design finds only preliminary direct evidence for computational thinking teaching support

Article Title: A systematic literature review of generative AI support for pre-service teachers in designing computational thinking instruction

Article References: Farah, H. L., & Budiyanto, C. W. (2026). A systematic literature review of generative AI support for pre-service teachers in designing computational thinking instruction. Discover Education, 5(1), Article 1210. https://doi.org/10.1007/s44217-026-02322-y

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02322-y

Keywords: generative AI, computational thinking, instructional design, pre-service teachers, systematic review, teacher education, systematic, literature, review, generative, support, pre-service