This study examines whether the sociopolitical contexts and regulatory environments of generative AI developers influence the outputs produced by these systems, with implications for research integrity. The research adopts a qualitative, comparative design, analysing responses from three generative AI models: ChatGPT (OpenAI, USA), Alice (Yandex, Russia), and DeepSeek (High-Flyer, China).
A structured prompt set consisting of 24 research questions was developed, focusing on politically and ideologically sensitive topics within political science and international humanitarian law (e.g., armed conflicts, sanctions, human rights accountability). Each prompt was submitted to each AI model three times at different intervals to test response consistency and reduce output variability, yielding a dataset of 216 AI responses.
The outputs were analysed using a thematic qualitative coding framework, combining elements of critical discourse analysis and comparative content analysis. Responses were evaluated across four key dimensions: (1) tone (neutral, justificatory, critical), (2) narrative framing (alignment with identifiable geopolitical perspectives), (3) source attribution and diversity, and (4) response completeness or refusal patterns.
The findings reveal systematic disparities across the models. ChatGPT generally produced balanced but Western-leaning institutional narratives; Alice frequently reflected positions aligned with Russian state perspectives, while DeepSeek demonstrated selective engagement, with partial omissions or cautious framing on politically sensitive issues. Notably, certain prompts elicited refusals or constrained responses in specific models, indicating embedded content-governance mechanisms.
These divergences suggest that generative AI systems are not ideologically neutral tools but are shaped by the legal, political, and cultural environments in which they are developed and deployed. The study argues that uncritical reliance on such tools in academic research risks reproducing hidden biases, thereby undermining scholarly objectivity and research integrity. It calls for increased methodological transparency, critical engagement, and cross-platform verification when integrating generative AI into academic workflows.