The Use of Generative Artificial Intelligence in the Digitisation of Printed and Manuscript Documents and Its Contribution to Historical and Archival Education

Objectives: This study examines how contemporary generative language models can support archival and historical work with Czech handwritten texts, focusing on transcription and basic preliminary analysis, and it outlines key limitations and ethical implications for educational use in archival science and digital humanities. Methods: A qualitative case study was conducted using seven modern personal handwritten Czech texts from the 1980s and 1990s (a poem written by a child, love poems, a school test, and study notes). Three widely available tools in their free versions (ChatGPT, Claude, and Copilot) were tested using identical task instructions. Outputs were comparatively evaluated with regard to transcription accuracy, content and stylistic interpretation, and recognition of selected formal features of the texts. The empirical comparison was complemented by a critical review of relevant scholarly literature and reflection on authenticity, data integrity, epistemic security, and personal data protection. Results: Claude achieved the best overall performance, followed by ChatGPT, while Copilot produced substantially weaker results in the tested tasks. Across tools, interpretation and analysis proved more challenging than transcription, and outputs included errors and over-interpretations that require expert verification. Conclusions: Generative language models can function as supportive tools for transcription, preliminary analysis, and didactic work, but they cannot replace professional archival or historical expertise. Responsible use requires critical human supervision and explicit attention to ethical and data-protection considerations.