The propagation of deepfakes across social and digital media presents significant challenges for media literacy education, as undergraduate students often exhibit low detection accuracy despite high confidence. This confidence-competence gap is particularly concerning for future K-12 teachers, who will be responsible for helping their own students in navigating an increasingly polluted digital information landscape. Current detection solutions include human-centered critical thinking and automated AI-based tools. However, over-reliance on automation may lead to cognitive offloading and in consequence, to declining critical thinking skills. For future teachers, maintaining strong analytical skills is paramount for effective media literacy education. Therefore, this research adopts a critical, human-centered approach to understanding cognitive strategies used by undergraduate teacher training students. We investigate whether their methods align with prior research and exceed random chance accuracy. We conducted qualitative semi-structured interviews with ten undergraduate students from the University of Fribourg’s teacher training program. During 40-minute sessions, students performed image classification tasks, distinguishing AI-generated from authentic images. We documented preferred cognitive strategies (intuitive vs. elaborate) by analyzing speech recorded during both intuitive and elaborate judgment phases. We examine how these strategies align with the Elaboration Likelihood Model and explore the impact of third-person effect theory and ego-involvement on classification accuracy. We emphasize developing and maintaining human-centered recognition strategies which strengthen individual critical thinking skills while remaining mindful of the potential risk that excessive cognitive offloading poses to conscious information evaluation, especially while navigating through contemporary information pollution on social media.

This study addresses three research questions:
RQ1: What practical deepfake detection strategies can be transferred to K-12 educational contexts?
RQ2: How do different reasoning strategies (intuitive vs. elaborate) impact identification performance?
RQ3: What role does ego-involvement play in classification accuracy?

Initial analysis reveals diverse cognitive strategies with promising intuitive and elaborate approaches. By the conference date, final findings on strategy effectiveness and ego-involvement's role in detection accuracy will be presented. Expected outcomes include evidence-based recommendations for enhancing critical thinking in media literacy education and practical strategies for integrating deepfake detection in teacher training programs. This study has two main limitations: a small sample size (ten students) limiting generalizability, and exclusive focus on image-based deepfakes. Future research should include larger samples and additional modalities (text, audio, video) to confirm results and provide more comprehensive understanding.

Keywords: Deepfakes, critical thinking, digital media literacy, teacher education.