A ninth-grade student submitted a well-structured essay analyzing symbolism in a novel. The paragraphs were coherent, the argument was focused, and the citations were correctly formatted. When the teacher asked the student to explain the central claim in their own words, the student went quiet. They had no memory of forming the argument, because they had not formed it.
The difference between having an answer and building one
When a student uses an AI tool to generate a response, they skip the part of learning that actually encodes knowledge. Decades of research on memory and cognition show that the effort of retrieving, organizing, and expressing information is not a side effect of learning. It is the mechanism. Cognitive scientists describe this as the "generation effect": information you produce yourself is retained far better than information you passively receive.
This matters because the visible output looks identical whether a student did the thinking or an AI did. A finished essay, a summarized reading, a completed problem set. The difference only surfaces later, in a conversation, on a follow-up assessment, or in a task that requires the student to actually apply what they supposedly learned. By that point, the gap between what was submitted and what was understood has already opened.
What the research says about unstructured AI use
A 2025 systematic review in Computers and Education examined how ChatGPT use affected critical and creative thinking in higher education. The review found a consistent pattern: when students used AI through structured activities that required them to explain their reasoning, compare approaches, or revise claims, their metacognitive behaviors improved. When they used AI without those structures, the benefits disappeared. The risk, the researchers concluded, is not AI use itself. It is unstructured, unscaffolded use that lets students hand off the cognitive work entirely.
A parallel review on NIH PubMed Central, examining AI scaffolding in STEM classrooms over 20 years, found the same boundary. Guided prompting, comparative analysis, and reflective writing helped students monitor their own thinking and develop what the researchers called "critical skepticism". Unstructured use, by contrast, carried the risk of outsourced cognition: students experience the appearance of learning without its substance. The difference between those two outcomes is almost entirely determined by how the teacher sets up the task.
Where teachers are seeing this first
The challenge is not that students are cheating in the traditional sense. Many students who use AI to draft their work do not think of it as dishonest. They see it as efficient. The distinction between using a tool and outsourcing the thinking is not always obvious to a fifteen-year-old, and it is not always articulated clearly by the adults around them.
Teachers who assign extended writing, argumentation, or analysis tend to notice the pattern first. A student submits strong written work but cannot speak to it in class. A student's written vocabulary is inconsistent with their verbal responses. ASCD's writing on designing for thought in AI-rich environments makes the point plainly: educators need to intentionally protect space for students to explain their reasoning, not just produce outputs. The output is not the goal. The thinking behind it is.
Practical ways to keep the thinking in the room
The most direct intervention is to add a verbal layer to any written task. If students submit work, ask them to explain their argument in two minutes without looking at it. This is a comprehension check, not a punishment. It shifts the incentive: submitting work you cannot explain stops being useful, because the next step requires you to understand it.
Oral assessment is one of the most reliable ways to verify that students have actually done the thinking. A structured one-on-one conversation, a class discussion with follow-up questions, or a recorded explanation gives you a window into student understanding that written outputs cannot provide on their own. It also signals clearly to students that the thinking matters, not just the paper they hand in.
If you want a practical way to conduct oral assessments at scale and make oral comprehension a regular part of your teaching, see how ArticulAI works.