Ideological encoding in generative AI systems: a critical discourse analysis with implications for teaching, learning, and assessment

This speculative study investigates how epistemological assumptions manifest in generative AI platforms when responding to ideologically charged questions, examining implications for teaching, learning, and assessment. We presented identical prompts on diversity, equity, and inclusion to models from OpenAI, Google, DeepSeek, Anthropic, and xAI. Using Fairclough’s three-dimensional critical discourse analysis framework, we examined textual features, discursive practices, and sociocultural practices in unedited responses. Findings reveal consistent discursive patterns across platforms, reflecting epistemologies grounded in pragmatic neoliberalism and instrumental rationality that emphasise solutionist, technocratic framings aligned with progressive academic discourse. Secondary meta level analysis revealed striking asymmetries: whilst all platforms readily generated progressive critiques, conservative perspectives emerged only when explicitly requested. This convergence across competing corporations suggests epistemological homogeneity stems from shared developmental contexts rather than platform specific design. As generative AI becomes integral to educational practice, these findings raise critical concerns about knowledge diversity in AI mediated learning environments. When students learn through systems encoding particular ways of knowing as natural, implications extend beyond content bias to encompass how learners understand knowledge itself. The study demonstrates an urgent need for critical AI literacy education helping students, teachers, and institutions recognise these systems as active participants in knowledge construction. We argue that educators must position themselves as essential mediators, cultivating approaches that foreground epistemological awareness and promote genuine pluralism in democratic education.