Meta-prompting strategy is most effective at improving academic text readability in ChatGPT compared to standard, roleplay, and chain-of-thought approaches.
Objective: To evaluate which ChatGPT prompting strategies are most effective at improving academic text readability, particularly for Swedish texts.
Methods:
- Analyzed 136 academic texts using four prompting approaches with ChatGPT-4
- Used LIX readability index to measure text complexity
- Tested Standard, Meta, Roleplay, and Chain-of-Thought prompting strategies
- Evaluated readability scores, text characteristics, and improvement metrics
Key Findings:
- Meta-prompting achieved the best overall readability improvements
- 91.91% of Meta-prompted texts reached "easy" reading level
- Meta approach produced texts with fewest long words
- Standard prompting showed inconsistent results, sometimes increasing text difficulty
- Meta and Chain-of-Thought strategies significantly outperformed Standard and Roleplay
Implications:
- ChatGPT can effectively simplify academic texts for diverse learning needs
- Different prompting strategies should be used based on specific readability goals
- Results support using AI tools to improve educational text accessibility
- Findings particularly relevant for languages with fewer simplification resources
Limitations:
- Study focused only on Swedish language texts
- Maximum 30 concurrent users recommended
- Simplified metrics for measuring readability
- Results may not generalize to other languages or contexts
Future Directions:
- Test effectiveness across multiple languages
- Evaluate long-term impact on student comprehension
- Develop teacher training materials and lesson plans
- Explore integration with pedagogical frameworks
Title and Authors: "Got It! Prompting Readability Using ChatGPT to Enhance Academic Texts for Diverse Learning Needs" by Elias Hedlin, Ludwig Estling, Jacqueline Wong, Carrie Demmans Epp, and Olga Viberg
Published On: March 2025
Published By: LAK 2025 (Learning Analytics and Knowledge Conference)