Researchers assign keywords to data to describe the observed or modeled physical properties. To ensure the discoverability and interoperability of this metadata, the keywords should be machine-readable and conform to standardized vocabularies or ontologies. The I-ADOPT framework provides guidelines for the formulation of such keywords in accordance with FAIR principles; however, the transformation of commonly used terms into atomic I-ADOPT components is still a highly manual task requiring both semantic and domain knowledge. In response, we propose an LLM-based workflow to generate FAIR-compliant descriptions of variables that conform to the I-ADOPT framework.