From Data to Decision-Making: A Review of Large Language Model Applications Across the Full Lifecycle of Geothermal Resource Development

Geothermal resource development involves heterogeneous geological and engineering records, uncertain reservoir parameters, and decisions constrained by physical processes and safety requirements. This narrative and critical review examines how large language models (LLMs), including retrieval-augmented and tool-assisted systems, could support knowledge-intensive tasks throughout geothermal exploration, development, and operation. It distinguishes direct geothermal evidence from empirical evaluations in adjacent fields and from conceptual proposals; results from petroleum engineering, mineral exploration, and broader energy systems are not treated as proof of geothermal performance. Peer-reviewed studies demonstrate useful capabilities in technical document extraction, domain question answering, knowledge-graph reasoning, and interactions with scientific software, but direct geothermal engineering validation remains limited. For reservoir inversion, drilling-risk warning, and design optimization, quantitative assessment must rely on measured data and validated physical or numerical models, with LLMs acting as auditable interfaces and information-integration tools. The review proposes a task-specific benchmark framework and examines cross-domain transfer, multimodal alignment, knowledge-graph and retrieval collaboration, uncertainty quantification, and LLM–physical-model coupling. These frameworks are literature-informed proposals rather than implemented geothermal systems. Key priorities include independent geothermal test sets, expert-annotated reference answers, calibrated uncertainty, physics-based verification, and transparent reporting of model failure. The resulting synthesis identifies credible current applications and the evidence requirements for reliable future deployment.