The rapid evolution of web technologies has brought both increased complexity and growing responsibility for ensuring digital inclusion. As dynamic front-end frameworks and multimedia content become ubiquitous, traditional accessibility validation methods often fall short. Recent advances in Large Language Models (LLMs) open new avenues for supporting web developers in identifying and mitigating accessibility barriers. This paper presents an empirical study evaluating the capabilities of LLMs to assess and improve the accessibility of web applications, including dynamically generated content. Our findings show that LLMs can provide valuable real-time feedback, especially in interpreting asynchronous behavior and reasoning about user interface accessibility. However, we also identify critical limitations. This study contributes to the broader effort of leveraging AI to create a more inclusive and equitable Web for all.
Andruccioli et al. (Wed,) studied this question.