Every day, an educational publishing house produces hundreds of thousands of images for e-books. Ensuring that these images are accessible to all students, including those with visual impairments, is not only an educational necessity but also a legal obligation under the European Accessibility Act (EAA). This paper presents an experiment that integrates Artificial Intelligence (AI) into the automatic generation of alternative text (alt text) for editorial images. The AI system is capable of analyzing visual content, interpreting its meaning, and generating coherent, context-aware alt texts that contribute to a more inclusive learning experience. Through a comparative analysis of various AI models, this paper highlights key benefits such as scalability, speed, and consistency in the production of alt text—without compromising semantic quality. However, the study also emphasizes the importance of maintaining a human in the loop approach. Human oversight remains essential. AI can streamline the process, but human expertise is critical to achieving truly accessible and meaningful educational content. The experiment thus represents a replicable best practice, supporting publishers in both regulatory compliance and social responsibility.