print

International political bias in large language models:
a critical discourse analysis of narratives in ChatGPT, LLaMA, Gemini, and DeepSeek

Zinovyeva E.S.,

MGIMO University, Moscow, Russia, elena.zinovjeva@gmail.com

elibrary_id: 625192 |

Trapeznikov V.P.,

MGIMO University, Moscow, Russia, traper274@gmail.com

elibrary_id: 1204189 |

Article received: 2025.07.10 20:59. Accepted: 2025.11.14 21:00


DOI: 10.17976/jpps/2026.01.11
EDN: VAVKWA

Rubric: Theory and methodology of political studies

For citation:

Zinovyeva E.S., Trapeznikov V.P. International political bias in large language models: a critical discourse analysis of narratives in ChatGPT, LLaMA, Gemini, and DeepSeek. – Polis. Political Studies. 2026. No. 1. https://doi.org/10.17976/jpps/2026.01.11. EDN: VAVKWA (In Russ.)

Abstract

This article explores the issue of international political bias in large language models (LLMs), using ChatGPT, Llama, Gemini, and DeepSeek as case studies. The research examines how these models construct assessments of key geopolitical actors–namely, the United States, Russia, China, Iran, and Israel– through the combined methodological lens of critical discourse analysis (CDA) and the emotional turn in International Relations (IR) theory. The study employs both automated sentiment analysis (via the NRC Emotion Lexicon) and qualitative discourse analysis of LLM-generated texts. The findings reveal a persistent structural asymmetry in the representation of states, which correlates with dominant narratives of Western political elites: the United States and China are predominantly associated with positive emotions (trust, joy), whereas Russia and Iran are more frequently linked to negative emotional frames (anger, fear). Although Israel’s foreign policy is subject to some criticism, its portrayal remains considerably more neutral compared to that of Russia and Iran. Moreover, LLMs appear to adopt discursive strategies of soft power, promoting Western normative concepts such as human rights and a “rules-based international order,” while marginalizing alternative geopolitical perspectives. Notably, the analysis also uncovers instances of criticism directed at U.S. foreign policy, suggesting that such bias is not uniform and may vary across contexts. The study contributes to ongoing scholarly debates on the role of AI in public diplomacy, algorithmic power, and technological sovereignty, underscoring the need for regulatory mechanisms to mitigate ideological distortions embedded in the outputs of generative language models.

Keywords

large language models (LLMs), artificial intelligence, emotional turn in the IR theory, soft power, information policy, algorithmic power, technological sovereignty.

Дополнительные материалы

Приложение

Abid, A., Farooqi, M., & Zou, J. (2021). Persistent anti-Muslim bias in large language models. Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’21), 298-306. https://doi.org/10.1145/3461702.3462624

Acharya, A. (2017). After liberal hegemony: the advent of a multiplex world order. Ethics & international affairs, 31(3), 271-285. https://doi.org/10.1017/S089267941700020X

An, J., Huang, D., Lin, C., & Tai, M. (2024). Measuring gender and racial biases in large language models. arxiv.org. https://doi.org/10.48550/arXiv.2403.15281

Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016). Man is to computer programmer as woman is to homemaker? Debiasing word embeddings. arxiv.org. https://arxiv.org/pdf/1607.06520

Crawford, K. (2021). Atlas of AI: power, politics, and the planetary costs of artificial intelligence. New Haven: Yale University Press. https://doi.org/10.12987/9780300252392

Dijk, T.A., van (2017). Discourse and power. Palgrave Macmillan. https://doi.org/10.1007/978-1-137-07299-3

Exler, A., Schutera, M., Reischl, M., & Rettenberger, L. (2025). Large means left: political bias in large language models increases with their number of parameters. arxiv.org. https://arxiv.org/abs/2505.04393

Fairclough, N. (2013). Critical discourse analysis: the critical study of language. Routledge. https://doi.org/10.4324/9781315834368

Faulborn, D., Sen, I., Pellert, M., Spitz, A., & Garcia, D. (2025). Only a little to the left: a theory-grounded measure of political bias in large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, vol. 1 (pp. 31684-31704). Vienna: Association for Computational Linguistics. https://doi.org/10.18653/v1/2025.acl-long.1529

Feng, S., Park, C.Y., Liu, Y., & Tsvetkov, Y. (2023). From pretraining data to language models to downstream tasks: tracking the trails of political biases leading to unfair NLP models. arxiv.org. https://doi.org/10.18653/v1/2023.acl-long.656

Fulay, S., Brannon, W., Mohanty, Sh., Overney, C., Poole-Dayan, E., Roy, D., & Kabbara, J. (2024). On the relationship between truth and political bias in language models. In EMNLP 2024 (pp. 9004-9018). https://doi.org/10.18653/v1/2024.emnlp-main.508

Gover, L. (2023). Political bias in large language models. The Commons: Puget Sound Journal of Politics, 4(1), 11-22.

Inoshita, K., & Zhou, X. (2024). Sentiment bias and security analysis in training datasets of large language models. In IEEE International Conference on Big Data and Cloud Computing (BDCloud) (pp. 1-8). https://doi.org/10.1109/BDCloud63169.2024.00008

Kotek, H., Dockum, R., & Sun, D. (2023). Gender bias and stereotypes in large language models. Proceedings of the ACM Collective Intelligence Conference (CI ’23) (pp. 12-24). https://doi.org/10.1145/3582269.3615599

Li, B., Haider, S., & Callison-Burch, C. (2024). This land is your, my land: evaluating geopolitical bias in language models through territorial disputes. In NAACL 2024 (pp. 3855-3871). https://doi.org/10.18653/v1/2024.naacl-long.213

Mohammad, S., & Turney, P. (2010). Emotions evoked by common words and phrases: Using Mechanical Turk to create an emotion lexicon. In Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text (pp. 26-34).

Navigli, R., Conia, S., & Ross, B. (2023). Biases in large language models: origins, inventory and discussion. ACM Journal of Data and Information Quality, 15(2), 1-21. https://doi.org/10.1145/3597307

Nye, J.S. (2004). Soft power: the means to success in world politics. New York: Public Affairs.

Pacheco, A.G., Cavalini, A., & Comarela, G. (2025). Echoes of power: investigating geopolitical bias in US and China large language models. arxiv.org. https://arxiv.org/abs/2503.16679

Rettenberger, L., Reischl, M., & Schutera, M. (2024). Assessing political bias in large language models. Journal of Computational Social Science, 8. https://doi.org/10.1007/s42001-025-00376-w

Schramowski, P., Turan, C., Andersen, N., Rothkopf, C.A., & Kersting, K. (2022). Large pre-trained language models contain human-like biases of what is right and wrong to do. Nature Machine Intelligence, 4, 258-268. https://doi.org/10.1038/s42256-022-00458-8

Westwood, S.J., Grimmer, J., & Hall, A.B. (2025). Measuring perceived slant in large language models through user evaluations. Stanford Graduate School of Business Working Paper No. 4262. Wodak, R., & Meyer, M. (2015). Methods of critical discourse studies. Sage.

Nikonova, E.A. (2018). Specifics of the implementation of the influencing and manipulative function in a political English-language essay. Tomsk State Pedagogical University Bulletin, 7, 9-14. (In Russ.) https:// doi.org/10.23951/1609-624X-2018-7-9-14

Ushakin, S.A. (1995). Speech as political action. Polis. Political Studies, 5, 149-153. (In Russ.)

See also: