Benchmarking large language models for geolocating colonial Virginia land grants

Virginia's seventeenth- and eighteenth-century land patents survive primarily as narrative metes-and-bounds descriptions, limiting spatial analysis. This study systematically evaluates current-generation large language models (LLMs) in converting these prose abstracts into research-grade latitude/longitude coordinates. A digitized corpus of 5,471 Virginia patent abstracts (1695–1732) is released, with 43 rigorously verified test cases for benchmarking. Six OpenAI models across three architectures—o-series, GPT-4-class, and GPT-3.5—were tested under two paradigms: direct-to-coordinate and tool-augmented chain-of-thought invoking external geocoding APIs. Results were compared against a professional GIS workflow, Stanford NER geoparser, Mordecai-3 neural geoparser, and a county-centroid heuristic. The top single-call model, o3-2025-04-16, achieved a mean error of 23 km (median 14 km), a 67% improvement over professional GIS methods and 70% better than Stanford NER. A five-call ensemble further reduced errors to 19 km (median 12 km) at minimal additional cost (~USD 0.20 per grant). Paired Wilcoxon tests confirm ensemble superiority (W=629, p=0.03 vs. single-shot). A patentee-name redaction ablation slightly increased error (~9%), showing reliance on metes-and-bounds reasoning rather than memorization. The cost-effective gpt-4o-2024-08-06 model maintained a 28 km mean error at USD 1.09 per 1,000 grants, establishing a strong cost-accuracy benchmark. External geocoding tools offer no measurable benefit for this task. These findings demonstrate that LLMs can georeference early-modern records as accurately and significantly faster and cheaper than traditional GIS workflows, enabling scalable spatial analysis of colonial archives.