Dyslexia presents significant challenges in education for students worldwide. While assistive technologies have been used to enhance readability, no study has systematically evaluated the ability of Language Models (LMs) to generate dyslexia-friendly text aligned with established accessibility guidelines. This proof-of-concept study assesses three state-of-the-art LMs on their ability to identify and apply dyslexia-friendly text criteria. Our findings reveal that their knowledge is limited and poses potential risks. To address this, we introduce DysText, a novel metric that quantifies dyslexia-friendly text characteristics based on the British Dyslexia Association’s Dyslexia Style Guide. Results indicate that while LMs can enhance the dyslexia-friendliness of texts, their responses should not be blindly trusted, underscoring the need for further verification.