<p>The difficulty of collecting social conflict data has caused its study to focus on recent events predominantly in the West. This paper shows how to use artificial intelligence to generate social conflict data regardless of language, location, or date. Digitizing tertiary books, lightly cleaning their text representation, and submitting that text to a large language model (LLM) produces accurate date, location, and event descriptions. These capabilities and results are demonstrated with books on Latin America after 1492, Imperial Russia, and Tokugawa Japan. Using LLMs requires a larger fixed cost than working a team of research assistants, but it produces results more quickly for most sources. It is less accurate for the Tokugawa Japan source; whether it or human translation is cheaper depends on the cost of correcting the LLM’s work. These results represent the floor of accuracy for artificial intelligence, suggesting researchers should soon use it for creating social conflict data from tertiary sources.</p>