New research is challenging the assumption that artificial intelligence tools help students think more deeply, with several recent studies and preprints suggesting AI-assisted schoolwork can be associated with reduced cognitive engagement and what one team of researchers calls “metacognitive laziness.” The findings arrive as federal education officials have promoted AI as a means to sharpen analytical skills, setting up a tension between policy ambition and emerging evidence from early studies and classroom-oriented research.
Federal Policy Meets Conflicting Evidence
The U.S. Department of Education issued guidance on AI use in schools through a Dear Colleague Letter and a proposed supplemental priority published in the Federal Register with a public comment period. In its official announcement, the department framed AI as a tool for sharpening critical thinking, urging educators to adopt the technology thoughtfully rather than avoid it outright.
That framing assumed schools would pair AI tools with strong pedagogical design. But more recent research and preprints tell a more complicated story. Rather than prompting students to think harder, AI assistants appear to make it easier for many learners to think less, particularly when tasks involve writing and argumentation. The gap between the federal vision and measured classroom behavior raises a direct question: are schools equipping students with a thinking aid or handing them a shortcut that weakens the very skills it was supposed to build?
What “Cognitive Debt” Looks Like in Practice
A study informally known as “Your Brain on ChatGPT,” available as an online preprint, compared groups of participants writing essays under three conditions: with an AI assistant, with a search engine, or with no digital tools at all. The AI-assisted group showed lower engagement on the study’s reported measures than either comparison group. Researchers described this pattern as an accumulation of “cognitive debt,” a term meant to capture how outsourcing mental effort to an AI creates a deficit in the kind of sustained thinking that builds knowledge over time.
The full manuscript suggests the problem is not simply that students might produce weaker work in the moment. The concern is cumulative: repeated reliance on AI-generated drafts may train students to skip the difficult, generative phase of writing where ideas take shape. If a learner rarely struggles through structuring an argument, the underlying skills involved in analytical reasoning and synthesis may get less practice. That is the debt, and it compounds with each assignment in which the machine handles the heavy lifting.
In classroom terms, cognitive debt can look mundane. A student assigned to compare two historical events may paste the prompt into an AI tool, skim the answer, and submit a lightly edited version. The work is technically complete, but the student has not interrogated causal links, weighed competing interpretations, or decided which evidence matters. Over time, this pattern can normalize a shallow relationship to complex tasks, even for students who appear to be performing adequately on surface-level assessments.
Lower-Ability Students Bear the Steepest Cost
Concerns about cognitive offloading are especially acute for students who already struggle. A separate study in vocational education settings, described by its authors as moving “from co-design to metacognitive laziness,” reported that learners frequently used generative AI in ways that bypassed effortful thinking rather than supplementing it. According to the authors’ analysis, lower-ability students were more likely to offload core cognitive work to AI tools than their higher-performing peers, relying on the system for idea generation, organization, and even reflection.
This finding complicates a common argument in ed-tech circles: that AI can serve as an equalizer, giving struggling students access to the same quality of support that well-resourced learners receive from tutors or enrichment programs. If weaker students are the most prone to letting AI do the thinking for them, the technology could widen existing achievement gaps rather than close them. The equity promise of AI in education depends heavily on how the tools are implemented, and the default mode of most popular AI assistants does not include the kind of scaffolding that would prevent passive use.
For teachers, this dynamic is difficult to detect. A polished essay produced with heavy AI assistance may look indistinguishable from a carefully mentored draft, especially when time and class sizes limit opportunities for oral defense or process-focused feedback. Without explicit norms and structured activities that reveal how students arrived at their answers, the learners who most need practice in planning, monitoring, and revising their thinking may instead become the most adept at hiding behind fluent machine prose.
Design Choices That Could Prevent Harm
Not all the evidence points toward inevitable decline in thinking skills. An experimental study on cognitive forcing functions in AI-assisted writing found that interface and interaction design can meaningfully increase or reduce overreliance. In that experiment, researchers tested AI systems that required users to create step-by-step plans before receiving generated text; the work on execution plans showed that prompting users to articulate their approach reduced misplaced trust in the AI and improved users’ sense of their own critical engagement.
The implication is practical: the problem may not be AI itself but the way most tools are built. Chatbot-style systems typically default to generating complete answers on demand. A tool designed to ask students clarifying questions before offering suggestions, or to require an outline before producing a draft, could shift the dynamic from passive consumption to active collaboration. Teaching guidance from university-based resources echoes this distinction, noting that while reliance on AI can lead to superficial engagement with complex ideas, structured use can instead promote deeper analysis and metacognitive awareness.
That gap between what AI systems could do for learning and what they currently do in most classrooms is where schools now find themselves. Many classrooms still rely on consumer-grade chatbot tools that do not include built-in forcing functions or mandatory planning interfaces. The default classroom experience with AI remains the consumer-grade chatbot, optimized for speed and convenience rather than cognitive development. Unless educators and vendors deliberately redesign interactions around planning, explanation, and critique, the path of least resistance will continue to favor shortcuts over struggle.
A Baseline That Predates the AI Surge
One reason the current moment matters is that international benchmarks for student thinking skills were established just before generative AI entered widespread classroom use. The OECD’s PISA 2022 assessment, presented as a study of creative thinking across participating countries, created a baseline for how well 15-year-olds can generate, evaluate, and improve ideas. That snapshot predates the explosion of ChatGPT-style tools in schools, making it a valuable reference point for tracking whether creative and critical thinking scores shift in the years ahead.
Early adopters of AI in education will therefore be operating under an unusually clear measurement window. If future PISA cycles show declines in students’ ability to frame problems, propose original solutions, or refine their work in response to feedback, researchers will have both policy timelines and usage patterns to compare. Conversely, if scores improve in systems that have tightly integrated AI with structured pedagogy, that too will offer evidence about the conditions under which AI can support rather than erode higher-order skills.
The OECD’s digital outlook underscores how quickly these conditions are evolving, documenting rapid growth in AI-powered platforms and data-driven tools across education systems. As adoption accelerates, the window for careful experimentation may narrow: practices that begin as pilots can become entrenched norms long before their long-term cognitive effects are fully understood.
Balancing Innovation With Cognitive Integrity
For policymakers and school leaders, the emerging research does not mandate abandoning AI but does argue for a more guarded optimism than early rhetoric suggested. Federal guidance that emphasizes critical thinking will need to be backed by procurement standards, training, and classroom routines that actively resist cognitive offloading. That could mean favoring tools that embed planning steps, reflection prompts, or explanation requests, and designing assignments that require students to show their process as well as their final product.
At the classroom level, educators can treat AI as an object of inquiry rather than a black-box assistant. Asking students to critique AI-generated responses, compare them with their own drafts, or revise machine output with explicit rationales can turn a potential shortcut into a site of metacognitive practice. Such approaches do not erase the risks documented in recent studies, but they acknowledge a central lesson of the evidence so far: without intentional design and guidance, AI will default to saving students effort, not building their capacity to think.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
