A methodological protocol for the generation and evaluation of AI-generated cultural heritage content

The growing use of generative artificial intelligence (GenAI) in cultural heritage dissemination has outpaced the development of rigorous procedures to ensure historical accuracy, formal coherence, and educational reliability. This article proposes a methodological protocol for the generation and evaluation of AI-generated cultural heritage content, with a particular focus on image-based representations. To establish the need for such a protocol, a bibliometric review of literature published between 2020 and 2025 in Web of Science and Scopus was conducted, confirming the absence of a comprehensive and replicable framework specifically designed to assess the historical veracity of AI-generated heritage imagery. The proposed protocol integrates four components: (1) a standardised image-generation procedure based on zero-shot prompting and first-output selection; (2) an interdisciplinary expert evaluation instrument comprising 12 items organised into five analytical dimensions; (3) an automated multimodal validation layer designed to contrast human and AI-based assessments; and (4) a predefined analytical plan encompassing internal consistency, inter-rater agreement, descriptive statistics, and qualitative thematic analysis. Rather than presenting empirical findings, the article offers a reproducible framework intended to support future comparative studies and professional applications in heritage communication, education, and digital dissemination. This article offers a reproducible framework intended to support future comparative studies and professional applications in heritage communication, education, and digital dissemination. The proposed protocol aims to reduce the risks of misinformation, enhance methodological transparency, and promote historically grounded uses of GenAI in cultural heritage contexts.