Improving FAIRness of Geospatial data using Large Language Models.

The FAIR principles- Findability, Accessibility, Interoperability and Reusability have become central to enhancing the value, transparency, and longevity of scientific data. Yet in the geospatial domain, implementing FAIR practices often remains uneven, reflecting gaps between formal standards compliance and practical usability within complex data infrastructures. Here, we examine how large language models (LLMs) can support the operationalisation of FAIR-aligned geospatial data stewardship. Drawing on practical experiences in geospatial FAIR implementation and a targeted synthesis of emerging LLM-enabled applications, we discuss how LLMs can assist various FAIR-related elements, including metadata enrichment, semantic alignment, multilingual discovery, and natural-language interaction with geospatial services. We further address technical, institutional and epistemic challenges associated with using LLMs and propose solutions that include hybrid validation approaches, confidence-aware workflows and transparent AI provenance indicators to mitigate automation bias and support trustworthy reuse. Building on these insights, we outline a research agenda for AI-supported FAIR (fAIr) centred on evaluation benchmarks, domain-tuned and resource-efficient models, explainable and auditable AI, and standards-integrated system architectures. We frame LLMs as assistive technologies that can help scale and embed FAIR principles within evolving geospatial research infrastructures, rather than replacements for traditional stewardship practices.