Abstarct: Generative AI tools such as ChatGPT and Google Gemini are now widely used by EFL learners during academic writing, but frequent use and polished final texts do not necessarily indicate genuine cognitive engagement. This study investigated the Reasoning Gap — a proposed pattern in which learners report critical evaluation of AI feedback while acknowledging reduced thinking effort — among eight Iraqi university EFL students. A descxriptive-analytical mixed-methods design combined the 20-item bilingual AI Engagement Scale (AIES), covering Behavioural, Cognitive, Emotional, and Dependency engagement, with writing tasks and interviews for three focal participants. Findings showed high behavioural and emotional engagement (M = 4.40 each), alongside a lower, more variable cognitive-reflection item (Q9, M = 3.38). Six of eight participants (75.0%) showed a reasoning-gap indicator, pairing critical evaluation (Q7) with reduced thinking effort (Q19). The three case studies showed different behavioural forms of this pattern, from careful independent rewriting to near-verbatim copying and marked difficulty writing without AI. The study does not claim GenAI causes declining critical thinking; its exploratory sample precludes generalisation, but shows that high engagement can coexist with uneven reflection and reduced independent effort. EFL teachers therefore need process-oriented tasks, explicit AI-literacy instruction, and assessment that makes reasoning visible, rather than relying on the fluency of the final text alone.