From readable to executable

Abstract Traditional terminology standards often assume that trained users can resolve coexisting translation variants through contextual judgement. As generative artificial intelligence (genAI) and large language models (LLMs) increasingly mediate terminology reuse, this raises questions about whether standards encode sufficiently explicit term-selection information for auditable automated selection. This study uses ChatGPT-4o in a prompt-testing procedure to examine whether the WHO’s 2007 and 2022 Traditional Chinese Medicine (TCM) terminology standards provide sufficient guidance for AI-mediated selection among coexisting English variants. Each sampled term was tested using the same three-prompt sequence: constrained variant selection, brief justification, and a within-session shift from general medical use to possible use in a WHO terminology standard. The model consistently followed the single-choice restriction. Its rationales clustered into four observable justification patterns. Most responses to the third prompt stated that the initial choice would remain unchanged; these responses are reported only as statements produced under this specific sequence. The findings suggest that terminology standards designed for human readability may not provide sufficiently explicit guidance for AI-mediated variant selection. The study therefore proposes a governance stress-test procedure and a minimal set of candidate executable fields for term selection, subject to expert validation and controlled testing.