From Code to Conversation: What LLMs Mean for Political Science

Computational political scientists have used machine learning for decades. But LLMs bring qualitatively new capabilities such as vast context windows and a sophisticated understanding of natural language. This allows LLMs to help with a wide variety of tasks, such as information extraction, labelling, sentiment analysis, concept formation, and synthetic data creation. Anecdotal evidence suggests that LLMs have even become sophisticated enough to overrule those initially made by human experts (Ornstein et al. 2025: 272). While not without drawbacks such as biases engrained in models through training data (Motoki et al. 2024), LLMs open exciting novel research avenues (e.g., Halterman 2025; Meher and Brandt 2025; Steinert and Kazenwadel 2025). One theme of this Workshop is to contribute to the development of best practices for the application of LLMs in political science. In addition to questions of how to best employ LLMs, the debate must move beyond whether LLMs can help us. The 'utility' phase of adoption is already well underway across the social sciences. What now also comes into view are the deeper implications of what we might call the AI-turn in political science. How do LLMs reshape the very questions we ask, the kinds of knowledge we produce, and the power relations embedded in global politics and our own academic practices? By focusing on these dimensions, we move past initial fascination with LLMs as helpful tools to examine their broader impact on knowledge production, political practice, and planetary sustainability.

Halterman, A. (2025) ‘Synthetically generated text for supervised text analysis’, Political Analysis 33(3): 181–194. Le Mens, G. and Gallego, A. (2025) ‘Positioning Political Texts with Large Language Models by Asking and Averaging’, Political Analysis 33(3): 274–282. Leo, R. D., Zeng, C., Dinas, E. and Tamtam, R. (2025) ‘Mapping (A)Ideology: A Taxonomy of European Parties Using Generative LLMs as Zero-Shot Learners’, Political Analysis 33(4): 456–463. Meher, S. and Brandt, P. T. (2025) ‘ConflLlama: Domain-specific adaptation of large language models for conflict event classification’, Research & Politics 12(3): 20531680251356282. Motoki, F., Pinho Neto, V. and Rodrigues, V. (2024) ‘More human than human: measuring ChatGPT political bias’, Public Choice 198(1): 3–23. Ornstein, J. T., Blasingame, E. N. and Truscott, J. S. (2025) ‘How to train your stochastic parrot: large language models for political texts’, Political Science Research and Methods 13(2): 264–281. Steinert, C. V. and Kazenwadel, D. (2025) ‘How user language affects conflict fatality estimates in ChatGPT’, Journal of Peace Research 62(4): 1128–1143.

1: How can LLMs be used without sacrificing key tenets such as replicability, transparency, and accessibility?
2: How does the reliance on LLMs reshape research agendas in political science?
3: How do AI infrastructures shape global power, sovereignty, and governance?
4: How should we confront the environmental and material footprint of LLMs?
5: What impact do LLMs have on political science, and which ethical rules should govern their use?

1: Methodological uses of LLMs in political science: ensuring replicability and transparency
2: Rethinking evidence, inference, and theory-building in political science with LLMs
3: Case studies of LLM applications in political research: conflict, parties, and policy
4: Bias, fairness, and validity in political text generation and classification with LLMs
5: LLM development, governance, and their intersections with global power hierarchies
6: The environmental and resource costs of LLMs: climate politics and sustainable innovation
7: Ethical challenges of LLM adoption in political science research and publishing
8: Cross-disciplinary and regional perspectives on LLMs in knowledge and political practice