A preliminary evaluation of generative ai tools for blind users: usability and screen reader interaction

The increasing use of Generative Artificial Intelligence (GAI) tools such as ChatGPT, Copilot, Perplexity and Gemini opens up new possible scenarios for supporting work and everyday activities. For people who are blind, the usability of such tools through screen readers is crucial to ensure their use of such AI-based technologies. In this study, we explore the accessibility and usability of the interfaces of four popular AI-based tools via screen readers through a combination of semi-automated evaluations and inspections conducted by both sighted and blind accessibility experts and screen readers with more than 20 years of experience. Navigation, labeling of control elements, feedback mechanisms, and prompt handling were considered in the study. The results point to usability difficulties in all tools, particularly in navigation structure, clarity of feedback and interactive elements. Although this work empirically explores the accessibility of AI-based tools it brings out the first critical issues that deserve further investigation. However, they are based on a small group of experts and thus should be considered preliminary and useful for future studies.