Supporting Primary School Teachers in Developing English-Medium Science Materials Using Generative AI

Abstract

Developing English-medium Science materials for primary school students requires careful attention to language accessibility and subject accuracy. At primary school Labschool UNESA 2 Surabaya, initial coordination, teacher surveys, interviews, and reviews of existing materials highlighted the need to adapt Grade 3 Science texts to students’ language proficiency. A community engagement program was therefore conducted to support teachers in this process. Twelve teachers attended two face-to-face sessions on 21 November and 2 December 2025. Working in three groups, they used ChatGPT and MagicSchool.ai to revise three Science chapters, followed by readability checks and facilitator feedback. Changes were assessed using Flesch Reading Ease, Flesch-Kincaid Grade Level, and estimated CEFR levels. A post-program questionnaire gathered participants’ feedback. The revised texts showed improved readability. Average Flesch Reading Ease scores increased from 60.7 to 74.3, while Flesch-Kincaid Grade Level scores decreased from 8.7 to 5.5. Estimated CEFR levels shifted from B1+–B2 toward A2–B1 or B1. Teachers also produced revised reading texts, vocabulary sections, worksheets, and reusable prompts. Survey results showed that 83% rated their understanding of prompting as good, while 92% evaluated the program positively. However, neither student comprehension nor improvement in individual teacher competence was directly measured. Classroom testing is needed to assess the materials’ effectiveness.