FROM DETECTION TO REVISION:

Abstract

As generative AI becomes a routine part of undergraduate writing, students need instruction that moves beyond factual checking towards identifying and revising the assumptions embedded in fluent prose. This paper develops a pedagogy of bias-focused revision within the VERIFY framework: Verify Sources, Evaluate Logical Flow, Review Rhetorical Strategies, Identify Biases and Assumptions, Formulate Feedback, and Your Contribution. It draws on empirical research on cultural, demographic, and representational bias and uses a cognitive-linguistic comparison of fifteen matched essay pairs as a bounded classroom case rather than as evidence of the general prevalence of bias. The pairs consist of ChatGPT-generated argumentative essays and student revisions by Gulf Arabic-speaking EFL undergraduates responding to whether parents should influence their children's career choices. In this corpus, students did not simply copy or correct AI-generated prose; their revisions often altered how authority, autonomy, responsibility, and cultural normality were framed. On this basis, the paper proposes an initial contrastive modelling task followed by a five-step routine: mark exact wording, explain its framing effect, plan a targeted change, revise independently, and reflect on the decision. The routine is designed for two 50-minute sessions, can be compressed after students know it, and is supplemented by selective extensions rather than a fixed list of activities. The paper also offers an analytic rubric that links textual evidence, diagnosis, revision, contextual judgement, and documented decision-making. It argues that critical AI literacy should be assessed not only through students' ability to detect unreliable content, but also through their ability to justify, revise, and take responsibility for the values their final texts reproduce.