Round Table 1 — Title Missing

Submit the same query to different AI systems and you may get answers that vary substantially in length, structure, emphasis, and factual interpretation. For casual users this might be a curiosity — for researchers relying on AI-assisted transcription to access historical manuscripts they cannot read themselves, it is a methodological problem. This roundtable asks how AI tools should be evaluated and designed when their outputs function not merely as helpful suggestions but as the primary gateway to historical evidence. The discussion connects directly to Transkribus and digital humanities practice: when users cannot independently verify a transcription, what standards of reliability, precision, and transparency should we demand — and how do we build the critical digital literacy to ask those questions in the first place?