A multi-method descriptive study of 113 future computer science teachers in Kazakhstan found that while students viewed ChatGPT favorably for code generation, debugging, and explaining programming logic, their self-reported documentation of verification and solution modification remained largely general. Conducted during a 16-week programming course at Kh. Dosmukhamedov Atyrau University, the investigation tracked how preservice educators integrated artificial intelligence into their coursework, their understanding of responsible tool use, and their subsequent academic performance.
To capture students’ perspectives on generative artificial intelligence, researchers administered the 20-item Prompt-Literacy Perceptions Questionnaire. The survey instrument demonstrated high overall internal consistency, yielding a Cronbach’s alpha of 0.94. However, the study authors emphasized that the questionnaire covered heterogeneous domains and was not subjected to factor-analytic or behavioral validation, meaning the scores reflect subjective perceptions and self-reported comprehension rather than an objective, validated measurement of prompt competence.
Throughout the 16-week instructional period, participants completed practical assignments alongside independent control tasks while submitting ChatGPT-use disclosure notes and reflective comments. The research team gathered multiple complementary data streams, combining course-grade records and code submissions with qualitative content analysis to observe recurring patterns in how preservice teachers described verifying, revising, and adjusting AI-assisted programming solutions.
Students consistently reported recognizing the necessity of evaluating model outputs and disclosing substantial AI assistance during practical work. Despite these stated ethical commitments, qualitative evaluation of student submissions revealed a noticeable gap in reporting depth, as students frequently provided general statements rather than granular, step-by-step accounts of how they verified or altered AI-generated code structures.
To evaluate potential code convergence among participants, investigators conducted pairwise code-similarity screening across student submissions. The screening showed that relatively few pairs of assignments exceeded the predefined 50 percent similarity threshold, indicating limited structural overlap among submitted solutions. Researchers cautioned, however, that code-similarity screening alone cannot establish whether individual students independently modified code received from ChatGPT.
An examination of academic performance showed that mean course scores remained numerically close across practical assignments and independent control tasks. Nevertheless, the researchers noted that differences in assessment conditions and grading criteria between these two assignment types preclude interpreting the similar scores as evidence of equivalent learning outcomes or equal underlying technical mastery.
Because the investigation employed a descriptive single-cohort design without a control group, the data were not analyzed to test participant-level associations or establish causal effects. Consequently, the findings characterize classroom practices and subjective outlooks within a single setting rather than demonstrating measurable improvements in students’ prompt-literacy skills over the instructional term.
The study highlights the pedagogical necessity of establishing explicit guidelines for artificial intelligence verification and transparent documentation within computer science teacher preparation. Emphasizing that favorable attitudes toward ethical tool use do not automatically translate into thorough verification practices, the researchers advocate for instructional methods that require students to make their reasoning, testing, and modification workflows fully visible.
The collaborative investigation brought together researchers across Kazakhstan, representing Kh. Dosmukhamedov Atyrau University, Yessenov University in Aktau, and K. Zhubanov Aktobe Regional State University. By situating the inquiry within a 16-week programming course, the cross-institutional team explored how future educators interact with generative artificial intelligence when tasked with solving programming problems, diagnosing software defects, and interpreting coding logic. The research framework focused on capturing students’ operational practices alongside their self-reported ethical perspectives, addressing how generative tools are integrated into pre-service teacher training.
To answer their central questions without relying on causal inferences, the investigators systematically compared quantitative indicators and qualitative evidence during final data interpretation. The quantitative data characterized general questionnaire trends, disclosure frequencies, pairwise code-similarity distributions, and academic grade records across the single cohort. In parallel, qualitative content analysis mapped recurring patterns in the types of assistance requested from ChatGPT, detailing how participants described reviewing, verifying, and modifying model-generated code. Comparing these distinct sources allowed the authors to evaluate whether students’ professed appreciation for responsible verification matched their documented implementation habits.
The authors also highlighted analytical constraints regarding construct validity and performance equivalence. Although the 20-item questionnaire showed strong overall internal consistency, its heterogeneous composition and absence of behavioral or factor-analytic validation mean scores reflect self-reported student perceptions rather than an objective, unidimensional measure of prompt literacy. Similarly, while students achieved numerically close scores across practical coursework and control assignments, differing evaluation criteria and test environments prevent concluding that learning outcomes were equivalent. Consequently, the study emphasizes establishing explicit instructional practices that require pre-service teachers to articulate their reasoning and verification workflows during programming education.
Subject of Research: Prompt literacy and verification practices of future computer science teachers using ChatGPT
Article Title: Prompt literacy of future computer science teachers in solving and debugging programming tasks with ChatGPT: perceptions, practices, and academic performance patterns
Article References: Rakhmetov, M., Kenzhegulov, B., Asset, T., Leskaliy, B., Saltanova, G., Turebayeva, K., & Muratkaliyeva, A. (2026). Prompt literacy of future computer science teachers in solving and debugging programming tasks with ChatGPT: perceptions, practices, and academic performance patterns. Frontiers in Education, 11, Article 1894455. https://doi.org/10.3389/feduc.2026.1894455
Image Credits: AI Generated
DOI: 10.3389/feduc.2026.1894455
Keywords: ChatGPT, prompt literacy, computer science education, preservice teachers, programming education, code debugging, generative AI, code generation, academic performance, code similarity screening, educational technology, Atyrau University, higher education, instructional technology, AI ethics in education, multi-method research, descriptive study
Cite Scienmag News
Courtney Benton. (November 21, 2026). Study Examines How Future Computer Science Teachers Use ChatGPT in Programming. Scienmag. https://scienmag.com/study-examines-how-future-computer-science-teachers-use-chatgpt-in-programming/
Courtney Benton. "Study Examines How Future Computer Science Teachers Use ChatGPT in Programming." Scienmag, 21 November 2026, https://scienmag.com/study-examines-how-future-computer-science-teachers-use-chatgpt-in-programming/. Accessed 21 November 2026.
Courtney Benton. "Study Examines How Future Computer Science Teachers Use ChatGPT in Programming." Scienmag. November 21, 2026. https://scienmag.com/study-examines-how-future-computer-science-teachers-use-chatgpt-in-programming/








