Conceptual divergence analysis: LLM-assisted mapping of a researcher’s conceptual vocabulary against their curated literature

Abstract Large language model (LLM)-based research tools often overlook a researcher-specific problem in manuscript writing: where does my use of a key theoretical term stand relative to the literature I want to address? We introduce Conceptual Divergence Analysis (CDA), an LLM-based retrieval-augmented generation pipeline to compare theoretical term usage in a researcher’s manuscripts with usage in a researcher-curated literature corpus. We propose reflective calibration as the individual-level scientometric function this workflow performs. We demonstrate CDA on two research programs, encompassing corpora of roughly 3,200 and 2,800 papers and four theoretical terms. The analyses reveal researcher-specific divergences consequential for theoretical positioning. In one case, the researcher draws a distinction the literature leaves underarticulated. In another case, CDA reveals how a term carries different meanings across disciplinary traditions, helping the researcher anticipate how different scholarly audiences may interpret its use. Register assignments are anchored to pre-computed clusters. An optional researcher-led validation step records the researcher’s judgment before revealing the model’s interpretation, making agreement or disagreement explicit. We release the complete pipeline as an open-source resource with a no-installation hosted demonstration on a public toy corpus, extending an individual-level tradition in scientometrics from citation structure to conceptual vocabulary.