A Comparative Analysis of ChatGPT,
DeepSeek, and Gemini Using Hofstede’s
Cultural Dimensions
PEER REVIEWED PAPER
By J A A P Jayasundrara
Sri Lanka
ABSTRACT
The increasing adoption of Large Language Models (LLMs) in professional environments has raised concerns about possible cultural biases in AI-generated recommendations. However, because LLM training data are often opaque, such claims remain difficult to verify directly. This study investigates whether ChatGPT, DeepSeek, and Gemini produce culturally distinct advisory patterns when generating project recovery recommendations across five national contexts: the United States, Japan, India, Sri Lanka, and the United Arab Emirates (UAE).
Using a controlled prompt-based experimental design, each model generated recovery advice for a two-week schedule delay under a fixed deadline. The study produced 45 independent responses (3 models × 5 countries × 3 rounds). Responses were analysed using Hofstede’s Individualism versus Collectivism (IDV) and Uncertainty Avoidance (UAI) dimensions. A five-point Likert scale with predefined coding indicators was used for consistency.
ChatGPT emphasised direct accountability and individual leadership (average IDV = 4.2, UAI = 3.4). DeepSeek showed collectivist and highly structured patterns (average IDV = 2.1, UAI = 4.3). Gemini produced moderate, context-adaptive responses (average IDV = 3.1, UAI = 3.7). Cultural mismatches were observed in South Asian contexts, where AI recommendations appeared more individualistic and risk-averse than local workplace norms typically suggest. The paper concludes that project managers should apply culturally aware prompt engineering and human validation when using AI-assisted recovery advice. Because commercial LLMs are continuously updated, the outputs captured here represent a time‑specific behavioural snapshot.
Keywords: cultural framing, large language models, project management, Hofstede dimensions, AI governance, generative AI
- INTRODUCTION
Project management practice is increasingly augmented by Generative Artificial Intelligence (GenAI). Large Language Models (LLMs) such as ChatGPT, DeepSeek, and Gemini are now widely used by professionals for schedule recovery, stakeholder communication, risk assessment, and decision support. While these tools provide rapid and accessible guidance, concerns have emerged regarding the extent to which AI-generated recommendations may embed implicit cultural assumptions.
Because LLMs are trained on large-scale internet-based datasets, their outputs may unintentionally reproduce dominant managerial ideologies, communication norms, and organizational practices originating from particular cultural environments. Consequently, recommendations generated by AI systems may not always align with the communication expectations and leadership norms of all national contexts.
This issue is particularly important in international project environments, where leadership expectations, conflict management approaches, and stakeholder communication styles vary significantly across cultures. A recommendation that appears effective in a highly individualistic workplace may be interpreted as overly aggressive or culturally inappropriate in collectivist or consensus-oriented environments. Culturally misaligned project recovery strategies may therefore reduce stakeholder trust, weaken team cohesion, and negatively affect project outcomes.
Previous research has examined cultural values in LLM outputs and broader forms of algorithmic bias. However, limited work has explored how AI systems frame project recovery recommendations across multiple national contexts using a standardized project management scenario. This study addresses that gap by investigating whether culturally distinct advisory patterns emerge when identical project recovery situations are presented to different commercial LLMs.
More…
To read entire paper, click here
How to cite this paper: Jayasundara, J. A. A. P. (2026). Cultural Framing in AI-Generated Project Recovery Advice: A Comparative Analysis of ChatGPT, DeepSeek, and Gemini Using Hofstede’s Cultural Dimensions; PM World Journal, Vol. XV, Issue VI, June. Available online at https://pmworldjournal.com/wp-content/uploads/2026/06/pmwj165-Jun2026-Jayasundara-Cultural-Framing-in-AI-Generated-Project-Recovery-Advice.pdf
About the Author
J A A P Jayasundara
Sri Lanka
J A A P Jayasundara is an undergraduate in the Department of Physical Sciences and Technology, Faculty of Applied Sciences, Sabaragamuwa University of Sri Lanka. His research interests include project management, artificial intelligence, cross-cultural communication, and the societal impact of large language models. He is located in Kurunegala, Sri Lanka and can be contacted at pathumjayasundara04@gmail.com.
