LLMs meet political concept formation: Addressing the genetic gap

The artificial intelligence (AI) revolution offers new challenges for political science. Data is growing on an unprecedented scale, and search across expanding knowledge spaces has made vast amounts of information and answers instantly available. Yet research quality still depends on the foundational work of concept formation. Access to “limitless knowledge” becomes productive only when integrated with disciplined forms of reasoning. This article identifies a structural misalignment in the current landscape—what we call a “genetic gap”—between the computational logic of Large Language Models (LLMs) and the concept-building rules of the social sciences. To address this gap, the article integrates a well-established methodology for building political science concepts—Hyperpolitics—with a Retrieval-augmented generation-LLM, outlining an original research protocol tailored to the emerging landscape. To demonstrate the value of this integration, the article applies AI-assisted analysis to examine literature on the Italian Council of State, showing how matrix-based constraints enable disciplined concept formation.