<p>The increasing use of large language models (LLMs) in higher education raises critical concerns about fairness, particularly in multilingual professional learning environments. This study investigates whether and how linguistic bias manifests in LLM-generated academic content, focusing on English–Hindi outputs within the context of MBA education. Using a multi-dimensional empirical framework, the study assesses linguistic bias not only through surface-level textual differences but also through deeper variations in content quality and thematic structure. Drawing on a prompt-based experimental design, responses generated by an LLM were systematically evaluated across core MBA domains. The analysis integrates quantitative linguistic measures (readability, response length, and sentiment), expert-based qualitative assessments (accuracy, completeness, originality, and fluency), and thematic modeling to examine how knowledge is structured across languages and disciplines. The findings reveal a pronounced and asymmetric linguistic bias: English outputs are consistently longer, more complex, and more formal, while Hindi outputs are shorter and more accessible but often less detailed. Domain-level analyses further show that linguistic complexity and originality vary across MBA subjects, with technical fields exhibiting greater density and lower novelty. Thematic analysis indicates that while LLMs produce coherent domain-aligned content, they tend to reproduce established disciplinary and linguistic hierarchies rather than fostering integrative perspectives. By offering a multi-dimensional assessment of linguistic bias in LLM outputs, this study contributes empirical evidence to ongoing debates on AI fairness in higher education and highlights the need for context-aware evaluation frameworks to support equitable and pedagogically appropriate AI deployment.</p>