Iterative Prompt Refinement for Dyslexia-Friendly Text Summarization Using GPT-4o

Researchers developed an AI-powered text summarization system using GPT-4o to create dyslexia-friendly content for approximately 10% of the global population who struggle with reading fluency. The system successfully generates readable summaries for news articles within four attempts, achieving stable performance across 2,000 samples with readability scores meeting accessibility targets.

• →GPT-4o was used to create an iterative prompt refinement pipeline for dyslexia-friendly text summarization targeting 10% of the global population.
• →The system achieved readability scores of Flesch Reading Ease >= 90 for most summaries within four attempts on 2,000 news articles.
• →Composite scores combining readability and semantic fidelity ranged from 0.13 to 0.73 with typical values near 0.55.
• →The research establishes an empirical baseline for accessibility-driven NLP applications using large language models.
• →The study motivates further human-centered evaluation with actual dyslexic readers to validate real-world effectiveness.