Cultural bias and cultural alignment of large language models
Open accessThe DOI matches its Crossref record; 46 of the 46 references it cites were found in a registry or an index.
Abstract
Culture fundamentally shapes people’s reasoning, behavior, and communication. As people increasingly use generative artificial intelligence (AI) to expedite and automate personal and professional tasks, cultural values embedded in AI models may bias people’s authentic expression and contribute to the dominance of certain cultures. We conduct a disaggregated evaluation of cultural bias for five widely used large language models (OpenAI’s GPT-4o/4-turbo/4/3.5-turbo/3) by comparing the models’ responses to nationally representative survey data. All models exhibit cultural values resembling English-speaking and Protestant European countries. We test cultural prompting as a control strategy to increase cultural alignment for each country/territory. For later models (GPT-4, 4-turbo, 4o), this improves the cultural alignment of the models’ output for 71–81% of countries and territories. We suggest using cultural prompting and ongoing evaluation to reduce cultural bias in the output of generative AI.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures


Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“At the same time, the reproduction of biases and representational patterns embedded in training data raises concerns about whose knowledge, styles, and voices are normalised in AI-mediated learning contexts (Bender et al., 2021; Tao et al., 2024; Sterlie et al., 2024).”
From From Cognitive Necessity to Cognitive Choice: Higher Education Assessment and Learning in the Age of Generative AI (2026) · cites this paper as Tao et al., 2024
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Culture's Consequences: Comparing Values, Behaviors, Institutions, and Organizations across Nations
Bhagat, R. S., & Hofstede, G. (2002) · The Academy of Management Review · doi:10.2307/4134391 · Cited by 21,664
ChatGPT for Good? On Opportunities and Challenges of Large Language Models for Education
Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · doi:10.35542/osf.io/5er8f · Cited by 7,552
ChatGPT for good? On opportunities and challenges of large language models for education
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · Learning and Individual Differences · doi:10.1016/j.lindif.2023.102274 · Cited by 7,326 · cited 1×
“…Mistral, and LLaMA, which are trained on Internet-scale textual data to process text and produce human-sounding language, are increasingly used by people in all aspects of their life, including education [32], medicine and public health [13, 45], as well as creative and opinion writing [50, 29].”
BackgroundModernization, Cultural Change, and the Persistence of Traditional Values
Inglehart, R., & Baker, W. E. (2000) · American Sociological Review · doi:10.2307/2657288 · Cited by 4,524 · cited 1×
“Culture plays a major role in shaping the way individuals think and behave in their daily lives by embedding a pattern of shared knowledge and values into a group of people [27, 23, 39, 43].”
BackgroundLarge language models in medicine
Thirunavukarasu, A. J., Ting, D. S. J., Elangovan, K., Gutierrez, L., Tan, T. F., & Ting, D. S. W. (2023) · Nature Medicine · doi:10.1038/s41591-023-02448-8 · Cited by 3,944 · cited 1×
“…Mistral, and LLaMA, which are trained on Internet-scale textual data to process text and produce human-sounding language, are increasingly used by people in all aspects of their life, including education [32], medicine and public health [13, 45], as well as creative and opinion writing [50, 29].”
Background(2005) · Cited by 3,670
Culture, dialectics, and reasoning about contradiction.
Peng, K., & Nisbett, R. E. (1999) · American Psychologist · doi:10.1037//0003-066x.54.9.741 · Cited by 1,755 · cited 1×
“Cultural differences also influence causal attributions of behavior, such as explaining others’ actions based on their individual traits versus situational factors [11], and human judgment, such as resolving contradictions through compromise versus logical arguments [40].”
BackgroundA Brief Overview of ChatGPT: The History, Status Quo and Potential Future Development
Wu, T., He, S., Liu, J., Sun, S., Liu, K., Han, Q.-L., & Tang, Y. (2023) · IEEE/CAA Journal of Automatica Sinica · doi:10.1109/jas.2023.123618 · Cited by 1,452 · cited 1×
“5-turbo incorporated Reinforcement Learning with Human Feedback (RLHF) [49].”
MethodPersonality and Culture Revisited: Linking Traits and Dimensions of Culture
Hofstede, G., & McCrae, R. R. (2004) · Cross-Cultural Research · doi:10.1177/1069397103259443 · Cited by 1,397 · cited 1×
“…values which emphasize subjective well-being, or survival values which emphasize economic and physical security) have demonstrated national variation in personality [24], technological innovation [47], trust in automation [10], privacy concerns [48], and health behaviors and outcomes [35].”
BackgroundDoes culture influence what and how we think? Effects of priming individualism and collectivism.
Oyserman, D., & Lee, S. W. S. (2008) · Psychological Bulletin · doi:10.1037/0033-2909.134.2.311 · Cited by 1,385 · cited 1×
“Culture plays a major role in shaping the way individuals think and behave in their daily lives by embedding a pattern of shared knowledge and values into a group of people [27, 23, 39, 43].”
BackgroundThe influence of culture: holistic versus analytic perception
Nisbett, R. E., & Miyamoto, Y. (2005) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2005.08.004 · Cited by 1,076
Causal attribution across cultures: Variation and universality.
Choi, I., Nisbett, R. E., & Ara Norenzayan, A. (1999) · Psychological Bulletin · doi:10.1037//0033-2909.125.1.47 · Cited by 1,055 · cited 1×
“Cultural differences also influence causal attributions of behavior, such as explaining others’ actions based on their individual traits versus situational factors [11], and human judgment, such as resolving contradictions through compromise versus logical arguments [40].”
BackgroundArtificial intelligence and communication: A Human–Machine Communication research agenda
Guzman, A. L., & Lewis, S. C. (2020) · New Media & Society · doi:10.1177/1461444819858691 · Cited by 951 · cited 1×
“How language is produced and transmitted has changed drastically as a result of digital communication technologies and applications of artificial intelligence (AI) [20], especially emerging generative AI applications such as ChatGPT [2].”
BackgroundCultural variation in eye movements during scene perception
Chua, H. F., Boland, J. E., & Nisbett, R. E. (2005) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0506162102 · Cited by 891 · cited 1×
“Cultural differences influence foundational perceptual processes, such as whether objects are processed independently (analytic) or in relation to their context (holistic), and people’s capacity to ignore environmental cues when focusing on an object against a complex background [38, 30, 12].”
Culture, control, and perception of relationships in the environment.
Ji, L.-J., Peng, K., & Nisbett, R. E. (2000) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.78.5.943 · Cited by 803 · cited 1×
“Cultural differences influence foundational perceptual processes, such as whether objects are processed independently (analytic) or in relation to their context (holistic), and people’s capacity to ignore environmental cues when focusing on an object against a complex background [38, 30, 12].”
BackgroundChatGPT and the rise of large language models: the new AI-driven infodemic threat in public health
De Angelis, L., Baglivo, F., Arzilli, G., Privitera, G. P., Ferragina, P., Tozzi, A. E., & Rizzo, C. (2023) · Frontiers in Public Health · doi:10.3389/fpubh.2023.1166120 · Cited by 708 · cited 1×
“…Mistral, and LLaMA, which are trained on Internet-scale textual data to process text and produce human-sounding language, are increasingly used by people in all aspects of their life, including education [32], medicine and public health [13, 45], as well as creative and opinion writing [50, 29].”
BackgroundAI-Mediated Communication: Definition, Research Agenda, and Ethical Considerations
Hancock, J. T., Naaman, M., & Levy, K. (2020) · Journal of Computer-Mediated Communication · doi:10.1093/jcmc/zmz022 · Cited by 658 · cited 1×
“AI has become integrated into daily routines and affects the way people consume and produce language [22].”
BackgroundHow does culture influence innovation? A systematic literature review
Tian, M., Deng, P., Zhang, Y., & Salmador, M. P. (2018) · Management Decision · doi:10.1108/md-05-2017-0462 · Cited by 407 · cited 1×
“…values which emphasize subjective well-being, or survival values which emphasize economic and physical security) have demonstrated national variation in personality [24], technological innovation [47], trust in automation [10], privacy concerns [48], and health behaviors and outcomes [35].”
BackgroundUsing large language models in psychology
Demszky, D., Yang, D., Yeager, D. S., Bryan, C. J., Clapper, M., Chandhok, S., Eichstaedt, J. C., Hecht, C., Jamieson, J., Johnson, M., et al. (2023) · Nature Reviews Psychology · doi:10.1038/s44159-023-00241-5 · Cited by 369
Biases in Large Language Models: Origins, Inventory, and Discussion
Navigli, R., Conia, S., & Ross, B. (2023) · Journal of Data and Information Quality · doi:10.1145/3597307 · Cited by 364
Should ChatGPT be biased? Challenges and risks of bias in large language models
Ferrara, E. (2023) · First Monday · doi:10.5210/fm.v28i11.13346 · Cited by 282 · cited 1×
“Our evaluation paradigm can be used as a human-in-the-loop approach to guide improvements in cultural alignment for LLMs [16].”
MethodShould ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models
Ferrara, E. (2023) · arXiv · doi:10.48550/arxiv.2304.03738 · Cited by 282
Fitting In or Standing Out? The Tradeoffs of Structural and Cultural Embeddedness
Goldberg, A., Srivastava, S. B., Manian, V. G., Monroe, W., & Potts, C. (2016) · American Sociological Review · doi:10.1177/0003122416671873 · Cited by 248 · cited 1×
“This may have interpersonal and professional consequences by signaling a lack of cultural embeddedness within an organizational context or misrepresenting the user to their readers [19].”
BackgroundArtificial intelligence in communication impacts language and social relationships
Hohenstein, J., Kizilcec, R. F., DiFranzo, D., Aghajari, Z., Mieczkowski, H., Levy, K., Naaman, M., Hancock, J., & Jung, M. F. (2023) · Scientific Reports · doi:10.1038/s41598-023-30938-9 · Cited by 243 · cited 1×
“For instance, AI-generated response suggestions in chat or email applications influence not only communication speed, diction, and emotional valence, but also interpersonal trust between communicators [25].”
BackgroundSelf-citationHow language shapes the cultural inheritance of categories
Gelman, S. A., & Roberts, S. O. (2017) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1621073114 · Cited by 200 · cited 1×
“Culture is a way of life within a society that is learned by its members and passed down from generation to generation – language plays a central role in this process of cultural reproduction [18].”
BackgroundAssessing Cross-Cultural Alignment between ChatGPT and Human Societies: An Empirical Study
Cao, Y., Zhou, L., Lee, S., Cabello, L., Chen, M., & Hershcovich, D. (2023) · Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP) · doi:10.18653/v1/2023.c3nlp-1.7 · Cited by 163
Atari, M., Xue, M. J., Park, P. S., Blasi, D. E., & Henrich, J. (2023) · doi:10.31234/osf.io/5b26t · Cited by 151
Culture shapes the evolution of cognition
Thompson, B., Kirby, S., & Smith, K. (2016) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1523631113 · Cited by 141 · cited 1×
“Such small individual-level cognitive biases can accumulate over time to shape the broader cultural system [46].”
BackgroundPerils and opportunities in using large language models in psychological research
Abdurahman, S., Atari, M., Karimi-Malekabadi, F., Xue, M. J., Trager, J., Park, P. S., Golazizian, P., Omrani, A., & Dehghani, M. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae245 · Cited by 111
Bhagat, R. S. (2002) · Academy of Management Review · doi:10.5465/amr.2002.7389951 · Cited by 105
Having Beer after Prayer? Measuring Cultural Bias in Large Language Models
Naous, T., Ryan, M. J., Ritter, A., & Xu, W. (2023) · arXiv · doi:10.48550/arxiv.2305.14456 · Cited by 96
Mackenbach, J. P. (2014) · Health & Place · doi:10.1016/j.healthplace.2014.04.004 · Cited by 81 · cited 1×
“…values which emphasize subjective well-being, or survival values which emphasize economic and physical security) have demonstrated national variation in personality [24], technological innovation [47], trust in automation [10], privacy concerns [48], and health behaviors and outcomes [35].”
BackgroundDesigning Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs
Barocas, S., Guo, A., Kamar, E., Krones, J., Morris, M. R., Vaughan, J. W., Wadsworth, W. D., & Wallach, H. (2021) · Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3461702.3462610 · Cited by 72 · cited 1×
“This study contributes comprehensive, longitudinal, and theoretically-grounded evidence from a disaggregated evaluation of cultural bias for five of the most widely used LLMs to date.”
BackgroundProbing the psychology of AI models
Shiffrin, R., & Mitchell, M. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2300963120 · Cited by 70 · cited 1×
“Question: Taking all things together, rate how happy you would say you are.”
BackgroundBaby steps in evaluating the capacities of large language models
Frank, M. C. (2023) · Nature Reviews Psychology · doi:10.1038/s44159-023-00211-x · Cited by 60 · cited 1×
“The mechanisms underlying how humans and LLMs approach survey questions may differ significantly [44, 17].”
BackgroundChatGPT and the rise of semi-humans
Al Lily, A. E., Ismail, A. F., Abunaser, F. M., Al-Lami, F., & Abdullatif, A. K. A. (2023) · Humanities and Social Sciences Communications · doi:10.1057/s41599-023-02154-3 · Cited by 59 · cited 1×
“How language is produced and transmitted has changed drastically as a result of digital communication technologies and applications of artificial intelligence (AI) [20], especially emerging generative AI applications such as ChatGPT [2].”
BackgroundThe Effect of Culture on Trust in Automation: Reliability and Workload
Chien, S.-Y., Lewis, M., Sycara, K., Liu, J.-S., & Kumru, A. (2018) · ACM Transactions on Interactive Intelligent Systems · doi:10.1145/3230736 · Cited by 58 · cited 1×
“…values which emphasize subjective well-being, or survival values which emphasize economic and physical security) have demonstrated national variation in personality [24], technological innovation [47], trust in automation [10], privacy concerns [48], and health behaviors and outcomes [35].”
BackgroundViberg, O., Kizilcec, R. F., Jivet, I., Martínez‐Monés, A., Oh, A. H., Mutimukwe, C., Hrastinski, S., & Scheffel, M. (2024) · Computers in Human Behavior Reports · doi:10.1016/j.chbr.2024.100416 · Cited by 50
Self-citationThe Ghost in the Machine has an American accent: value conflict in GPT-3
Johnson, R. L., Pistilli, G., Menédez-González, N., Duran, L. D. D., Panai, E., Kalpokiene, J., & Bertulfo, D. J. (2022) · arXiv · doi:10.48550/arxiv.2203.07785 · Cited by 50 · cited 2×
“Considering that LLMs tend to be trained on corpora of text that overrepresent certain parts of the world, this widespread adoption raises a critical question of cultural bias, which can be hidden in the way LLMs generate and interpret language [31, 9, 41, 37, 14].”
BackgroundSocial norms and dishonesty across societies
Aycinena, D., Rentschler, L., Beranek, B., & Schulz, J. F. (2022) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2120138119 · Cited by 43 · cited 1×
“Please use a scale from 1 to 4, where 1 is Very happy, 2 is Quite happy, 3 is Not very happy, 4 is Not at all happy.”
MethodEVS (2022) · HAL (Le Centre pour la Communication Scientifique Directe) · doi:10.4232/1.14021 · Cited by 41
GPT-4 vs. GPT-3.5: A Concise Showdown
Koubaa, A. (2023) · doi:10.20944/preprints202303.0422.v1 · Cited by 37
Studying large language models as compression algorithms for human culture
Buttrick, N. R. (2024) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2024.01.001 · Cited by 33 · cited 1×
“We calculated the mean of the rescaled individual-level scores for each of the remaining 107 countries in each year’s survey, and then calculated the mean of the country-year-level scores for each country.”
MethodBridging large language model disparities: skill tagging of multilingual educational content
Kwak, Y., & Pardos, Z. A. (2024) · British Journal of Educational Technology · doi:10.1111/bjet.13465 · Cited by 25 · cited 1×
“This can improve cultural alignment [41, 34] but requires resources that render this approach accessible to only a few.”
BackgroundWhy Japan is building its own version of ChatGPT
Hornyak, T. (2023) · Nature · doi:10.1038/d41586-023-02868-z · Cited by 16 · cited 1×
“For example, AI Sweden released a Swedish version of GPT 1 , and the government of Japan started development of a Japanese version of ChatGPT to address cultural and linguistic bias [26].”
BackgroundWorld Values Survey (1981-2022). Trend File
Haerpfer, C., Inglehart, R., Moreno, A., et al. (2022) · Multilingual Matters (Channel View Publications) · doi:10.14281/18241.23 · Cited by 16
Cited by
Looking Beyond the Hype: Understanding the Effects of AI on Learning
Bauer, E., Greiff, S., Graesser, A. C., Scheiter, K., & Sailer, M. (2025) · Educational Psychology Review · doi:10.1007/s10648-025-10020-8 · Cited by 153
“For example, generated materials and other outputs may reflect cultural or gender biases (e.g., Kotek et al., 2023; Tao et al., 2024), and learners’ input language can lead to biased outputs, as LLMs may contain language and dialect prejudices (Hofmann et al., 2024).”
The Uneven Impact of Generative AI on Entrepreneurial Performance
Otis, N. G., Clarke, R. P., Delecourt, S., Holtz, D., & Koning, R. (2023) · doi:10.31219/osf.io/hdjpk · Cited by 146
Domain Specialization as the Key to Make Large Language Models Disruptive: A Comprehensive Survey
Ling, C., Zhao, X., Lu, J., Deng, C., Zheng, C., Wang, J., Chowdhury, T., Li, Y., Cui, H., Zhang, X., et al. (2026) · ACM Computing Surveys · doi:10.1145/3764579 · Cited by 141
The life cycle of large language models in education: A framework for understanding sources of bias
Lee, J., Hicke, Y., Yu, R., Brooks, C., & Kizilcec, R. F. (2024) · British Journal of Educational Technology · doi:10.1111/bjet.13505 · Cited by 119
“…Zhao et al., 2019), gender and ethnic stereotypes (Garg et al., 2018), gender neural words (Zhao et al., 2019), cultural biases (Durrheim et al., 2023; Swinger et al., 2019; Tao et al., 2024), and studies that trace training documents to identify the origin of such biases (Brunet et al., 2019).”
BackgroundSelf-citationAI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances
Agarwal, D., Naaman, M., & Vashistha, A. (2024) · arXiv · doi:10.48550/arxiv.2409.11360 · Cited by 99
Metadata
- Authors
- Tao, Y., Viberg, O., Baker, R. S., & Kizilcec, R. F.
- Affiliations
- Cornell University (US)
- KTH Royal Institute of Technology (SE)
- University of Pennsylvania (US)
- Corresponding author
- Kizilcec, René F
- Venue
- PNAS Nexus
- Publisher
- Oxford University Press (OUP)
- Published
- 2024-09
- Volume
- 3
- Issue
- 9
- Pages
- pgae346-pgae346
- ISSN
- 2752-6542 (ISSN-L)
- PMC
- PMC11407280
- OpenAlex
- W4402565126
- Semantic Scholar
- 5f8bf881c80125452e4a73ad51fdb2c72c65c551
- Cited by
- 383
- References
- 40
- Citations by year
- 2026: 213 · 2025: 154 · 2024: 15
- FWCI
- 158.32
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- PNAS Nexus (publishedVersion, cc-by-nc)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Funders
- Jacobs Foundation Research Fellowship and Digital Futures
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Social Sciences › Psychology › Social Psychology › Cultural Differences and Values
- Keywords
- cultural bias, cultural alignment, large language models, cultural values, GPT-4, generative AI, cross-cultural assessment, national survey data, cultural homogenization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.