ChatGPT-Assisted Lesson Planning for Children's STEAM Arts Education: An Experimental Study on Benefits, Challenges, Methods, and a Prompt Framework

On this page

Synthesis: Title: Luo and Tahir (2025) present an experimental study comparing traditional teacher-generated lesson plans with ChatGPT-assisted plans in children's STEAM arts education, finding that AI-assisted plans were evaluated as significantly higher quality (median 20.5 vs. 17.6, p = .002, large effect) by six expert professors. The study maps three methods of teacher–AI collaboration, documents the benefits and challenges of AI-assisted lesson planning, and contributes an iteratively refined prompt framework built on a "Role–Instructions–End Goal (RIE)" template plus a "four points and one line" optimization rubric.

Key Findings

Significant quality gains: ChatGPT-assisted lesson plans scored higher than traditional plans across all 13 teachers (ChatGPT-assisted range 18–24 vs. traditional 11–20), with a statistically significant within-subject difference on the Wilcoxon signed-rank test (median 20.5 vs. 17.6, z = 3.059, p = .002) and a large effect size (r = 0.883). Six professors' blind ratings showed good inter-rater reliability (ICC = 0.847).
Three methods of AI-assisted planning: teachers used ChatGPT either to fill content gaps in a self-outlined lesson (Method A, favored by 60%), to check and optimize a complete self-made plan (Method B), or to generate a plan they then adapted (Method C). Method A was recommended because it balances teacher autonomy with ChatGPT efficiency.
Benefits: teachers reported improved efficiency (faster topic selection, quicker high-quality content, reduced planning time), support for interdisciplinary integration (61% rated 4+), and creative inspiration, with ChatGPT serving chiefly as an inspiration and gap-finding tool rather than a source of fully original innovation.
Challenges: outputs often lacked personalization and practicality (idealized, competition-grade plans unworkable in daily classrooms), overlooked child-safety constraints (e.g., suggesting carving knives for young children), reflected Western-centric cultural bias, and image/resource generation was sometimes logically flawed or irrelevant.
Prompt framework: an improved ChatGPT prompt framework was developed for children's art teachers, built on a Role (R), Instructions (I), End Goal (E) template adapted from the RISEN framework, plus an optimization step using "four points and one line" evaluation criteria — standardized, practical, engaging, and complete, with an extension dimension. Teacher acceptance ratings averaged above 4 on a 5-point scale.
Key integration stages: the most valuable support came at generating lesson plans (Stage 2) and searching/generating multimedia resources (Stage 3), which are the most time-consuming parts of traditional planning.

What this means for practice

Instructors. Outline the lesson yourself first and use ChatGPT to fill content gaps (the study's Method A), rather than generating a full plan to adapt afterwards; teachers who worked this way kept pedagogical control while saving the most time on topic selection and content drafting.
Instructors. Screen generated activities against child-safety constraints before use — teachers reported outputs that were unsuitable for young children, for example practical suggestions involving carving knives.
Instructors. Concentrate AI use at the two stages teachers reported as most valuable and most time-consuming: generating the plan itself and searching for or generating multimedia resources, and validate image outputs, which were sometimes logically flawed or irrelevant to the topic.
Designers. Build the RIE (Role–Instructions–End Goal) prompt template and the "four points and one line" criteria (standardized, practical, engaging, complete, plus an extension dimension) into the tool's interface, since teachers rated their acceptance of the refined framework above 4 on a 5-point scale.
Faculty developers. Support art specialists in interdisciplinary planning rather than assuming STEAM fluency: only two of the study's 13 teachers had STEAM-related backgrounds, and teachers used the tool precisely because interdisciplinary content was their weakest area.

Limitations

Thirteen children's art teachers and six professors participated, all from China with experience in Chinese children's art education, so both the planning and the rating reflect a single national and cultural context — one the authors note ChatGPT itself handles poorly, since its predominantly Western training data produces internationalized content with little grasp of local custom.
Only two of the 13 teachers had STEAM-related backgrounds, so the sample cannot support strong claims about STEAM arts education specifically, even though that is the study's stated setting.
The lesson-plan sample is 25 plans (12 traditional, 13 ChatGPT-assisted): one teacher used ChatGPT daily and submitted only a single plan, leaving 12 usable pairs for the Wilcoxon comparison.
Benefits and challenges rest mainly on self-reported questionnaires and interviews, which the authors acknowledge may reflect socially desirable responding; evaluation was limited to professor rubric scores, and no student data were collected — the study never measured how the plans affected learners.

Connected Concepts

Connected Articles

Citation

Luo, Z., & Tahir, R. (2025). ChatGPT-assisted lesson planning for children's STEAM arts education: An experimental study on benefits, challenges, methods, and a prompt framework. International Journal of Artificial Intelligence in Education, 35, 3696–3745.

Metadata

Study design	Experiment
Discipline	Stem Education
Level	Preschool, Primary Education
Intended audience	Instructors, Curriculum Designers, Faculty Developers
Learning and instruction	Creativity
Technologies and techniques	Generative AI, Prompt Engineering
Assessment and measurement	Educational Measurement
Foundations of AI in education	Curriculum Design, Teaching

Source History Edit Chat