This study introduces an ongoing research project at the intersection between accessibility, Artificial Intelligence (AI) and publishing. The aim of this research is to develop an AI-driven methodology designed to streamline the process of alternative text (alt-text) generation for accessible eBooks, a task that has traditionally been both challenging and costly. This research seeks to fulfill a critical need within the publishing industry by offering an AI-based solution that harmonizes streamlined publishing processes with the requirements of digital accessibility standards and regulations. The research work follows a practical and results-oriented approach, and the study is structured around three main steps. The initial stage entails an examination of multiple GAI systems for alt-text production, alongside with the establishment of a metric for the evaluation of alt-text quality. In the second phase, an automated system is developed to extract images and related content from EPUB files; this system interfaces with the selected GAI tools and is tested on real-world cases. The third phase comprises the evaluation of the generated alt-texts, providing insights into the performance of the different tools tested. Once these three phases are completed, this research aims to provide publishers with an evidence-based methodology for implementing semi-automated alt-text generation, potentially transforming accessibility workflows in digital publishing. The proposed methodology advocates for a semi-automated approach to alt-text production. Although GAI technology facilitates efficiency and decreases production costs, the editorial knowledge is vital for the ultimate assessment of alt-text, guaranteeing that the results are of high quality and comply with accessibility requirements.