Generative AI (Gen-AI) is reshaping how students and early-career researchers discover, select, and cite scholarly sources. Large Language Models and AI-assisted discovery tools increasingly suggest references automatically, often prioritising accessibility and algorithmic relevance over disciplinary authority, peer review, and epistemic rigor. As a result, traditional citation instruction focused on formatting compliance and plagiarism prevention is no longer sufficient. This pedagogical article argues for a shift toward critical citation literacy. Rather than treating citation as a mechanical skill, this approach frames it as an epistemic and ethical practice. Drawing on scholarship in information literacy, academic integrity, and the sociology of knowledge, the article identifies key risks associated with Gen-AI, including fabricated references, citation without reading, misattribution, and the amplification of existing visibility biases. It differentiates between classes of AI tools, such as parametric language models and retrieval-augmented systems, to show that citation risks vary by infrastructure rather than treating “AI hallucination” as a uniform phenomenon. The article proposes a five-dimensional framework for teaching critical citation literacy and outlines practical teaching and assessment strategies. In AI-mediated research environments, citation education must move beyond rule enforcement toward cultivating epistemic responsibility, judgment, and accountability while safeguarding intellectual diversity.
Madsen et al. (Sun,) studied this question.