Exploring and Comparing Scaffolding Strategies of ChatGPT-3.5 and a Customized GPT for Reading Comprehension
Saved in:
| Title: | Exploring and Comparing Scaffolding Strategies of ChatGPT-3.5 and a Customized GPT for Reading Comprehension |
|---|---|
| Language: | English |
| Authors: | Udomsak Sirita, Richard Watson Todd |
| Source: | rEFLections. 2025 32(3):1793-1819. |
| Availability: | King Mongkut's University of Technology Thonburi School of Liberal Arts. 126 Pracha Uthit Road, Bang Mod, Thung Khru, Bangkok, Thailand 10140. Tel: +66-2470-8756; Fax: +66-2428-3375; Web site: https://so05.tci-thaijo.org/index.php/reflections/index |
| Peer Reviewed: | Y |
| Page Count: | 27 |
| Publication Date: | 2025 |
| Document Type: | Journal Articles Reports - Research |
| Education Level: | Higher Education Postsecondary Education |
| Descriptors: | Artificial Intelligence, Technology Uses in Education, Reading Comprehension, Foreign Countries, Undergraduate Students, Reading Tests, Multiple Choice Tests, Computer Assisted Testing, Scaffolding (Teaching Technique), Reading Instruction, Error Correction, Misconceptions, Critical Thinking, Instructional Effectiveness, Instructional Design |
| Geographic Terms: | Thailand |
| ISSN: | 1513-5934 2651-1479 |
| Abstract: | This study compares scaffolding strategies generated by ChatGPT-3.5 and a customized GPT in reading comprehension exercises to assist Thai university students in achieving a minimum CEFR B2 level as a requirement for Thai bachelor's degree programs. A prompt for ChatGPT-3.5 was designed to generate four reading passages, each with five multiple-choice questions. A similar approach was used to configure a customized GPT, also with a prepared file containing four reading passages and five multiple-choice questions. Data were collected based on the responses from both versions when two incorrect and one correct answer were selected respectively for each question. The results revealed that the customized GPT generated more meaningful and diverse scaffolding strategies, whereas ChatGPT-3.5 produced consistent but limited responses focused on specific reading strategies. Furthermore, the study found that some valuable strategies, such as misconception correction and the promotion of critical thinking, were absent in ChatGPT-3.5. While both versions offer educational value, they differ in the depth and range of scaffolds provided. Educators and researchers should carefully consider these differences when integrating generative AI into instructional design. In particular, this study highlights the importance of grounding AI-assisted learning in established pedagogical theories, such as scaffolding, to support core language skills like reading comprehension. As generative AI becomes more common in classrooms, thoughtful implementation and training for both instructors and students will be key to maximizing its potential within the Thai educational context. |
| Abstractor: | As Provided |
| Entry Date: | 2026 |
| Accession Number: | EJ1498344 |
| Database: | ERIC |
| Abstract: | This study compares scaffolding strategies generated by ChatGPT-3.5 and a customized GPT in reading comprehension exercises to assist Thai university students in achieving a minimum CEFR B2 level as a requirement for Thai bachelor's degree programs. A prompt for ChatGPT-3.5 was designed to generate four reading passages, each with five multiple-choice questions. A similar approach was used to configure a customized GPT, also with a prepared file containing four reading passages and five multiple-choice questions. Data were collected based on the responses from both versions when two incorrect and one correct answer were selected respectively for each question. The results revealed that the customized GPT generated more meaningful and diverse scaffolding strategies, whereas ChatGPT-3.5 produced consistent but limited responses focused on specific reading strategies. Furthermore, the study found that some valuable strategies, such as misconception correction and the promotion of critical thinking, were absent in ChatGPT-3.5. While both versions offer educational value, they differ in the depth and range of scaffolds provided. Educators and researchers should carefully consider these differences when integrating generative AI into instructional design. In particular, this study highlights the importance of grounding AI-assisted learning in established pedagogical theories, such as scaffolding, to support core language skills like reading comprehension. As generative AI becomes more common in classrooms, thoughtful implementation and training for both instructors and students will be key to maximizing its potential within the Thai educational context. |
|---|---|
| ISSN: | 1513-5934 2651-1479 |
