<![CDATA[Abstract A correct answer to a grammar question does not reveal whether a student reasoned through the item or relied on artificial intelligence. This qualitative study examined cognitive engagement in records submitted by 30 EFL students who completed five AI-assisted grammar items, producing 150 interaction episodes. Each episode contained a final answer, the prompt entered into an AI tool, and the copied AI response. The records were analyzed using the ICAP framework and an adapted coding rule focused on students’ visible contribution to grammatical content. All 150 episodes were coded as Passive; none provided the evidence required for Active, Constructive, or Interactive engagement. Although 24 prompts sought direct answers and 126 requested reasons, translations, or grammatical explanations, explanation-seeking had been modeled by the lecturer and the explanations were generated by AI. No episode contained an initial student answer for verification, student-written grammatical reasoning, or a substantive follow-up to the AI response. All final answers were correct, but the records could not establish independent understanding. The findings show that detailed prompts and extensive AI explanations should not be treated as learner-generated reasoning. Grammar tasks should require students to make their thinking visible before and after consulting AI. Keywords: AI-assisted grammar learning; cognitive engagement; EFL students; ICAP framework; student–AI interaction]]>
