This study evaluates the accessibility of multilingual websites of public universities in Romania. University websites play a central role for the academic community, and inaccessible content can exclude individuals with disabilities from participating in educational, professional, or social activities. The analysis focused on 32 homepages, considering English and French localised versions, with particular attention to the needs of visually impaired users. To assess web accessibility compliance with Web Content Accessibility Guidelines 2.1, we performed two tests: an automated evaluation to identify common language-related errors, and a complementary analysis using ChatGPT to examine the identified errors that typically require manual verification. Findings confirmed that the multilingual content on the tested university websites contains many language-related barriers. The AI-based evaluation proved effective when compared to manual assessment, as ChatGPT evaluated errors such as untranslated or redundant alternative text, mismatches between link descriptions and their purpose, untranslated page titles, or incorrect language declarations in 75% of cases. Positive outcomes suggest that this preliminary study could serve as a foundation for further research into the intersection of localisation, web accessibility, and artificial intelligence, moving beyond controlled tests.