A Major Investment in Educational Evidence
The Education Endowment Foundation (EEF) has launched a significant research commissioning call, committing up to £2.5 million to investigate the impact of generative artificial intelligence (GenAI) on student cognition and learning.
The initiative, announced on June 10, 2026, aims to build a robust evidence base as GenAI tools become an increasingly pervasive part of students' everyday learning. The research focuses specifically on school-aged learners in England, primarily those aged 13 to 15, where the use of these tools is most pronounced. According to recent data, two in three teenagers are already using GenAI to support their literacy and learning. Read about the rise of AI psychological support Explore how AI tutors transform education Understand GenAI safety frameworks
Research Objectives: Beyond Usage to Cognitive Mechanisms
The EEF’s call explicitly moves beyond documenting how students use GenAI to investigate why and how much it alters core cognitive processes. The central theoretical framework guiding this research is cognitive offloading—the phenomenon where individuals delegate mental tasks traditionally performed internally to external tools. This concept, grounded in decades of cognitive psychology (Risko & Gilbert, 2016), posits that while offloading can be adaptive, it can also undermine the development of enduring cognitive skills when it replaces the very processes that build expertise.
The EEF is particularly interested in whether GenAI is being used to offload processes critical for learning: recall, planning, reasoning, drafting, evaluation, and problem-solving. The research seeks to determine whether these offloading behaviors are temporary or if they lead to a persistent reduction in students’ capacity to engage in these activities independently.
The study’s primary cognitive outcomes include:
- Short-term and working memory: Can students retain information they’ve offloaded to AI?
- Attention and focus: Does reliance on AI reduce the ability to sustain focus on complex tasks?
- Critical thinking: Are students able to evaluate, critique, and synthesize information when AI generates it for them?
- Metacognition: Do students develop accurate self-awareness of their understanding? Or do they develop an "illusion of competence"?
- Knowledge acquisition: Is learning superficial, or is deep, transferable knowledge being built?
Secondary outcomes under investigation include motivation, self-efficacy, and resilience—factors that may be eroded when students perceive AI as a substitute for their own intellectual effort.
Critically, the EEF is not seeking descriptive studies. It is demanding experimental designs capable of generating robust causal evidence. Proposals must include a credible counterfactual—such as a control group using traditional methods or a group using AI under strict constraints—to isolate the causal impact of GenAI use. This focus on experimental rigor ensures the findings will be actionable for educators and policymakers.
The Cognitive Offloading Paradox: Immediate Gains, Long-Term Losses
The EEF’s research is being commissioned against a backdrop of emerging, alarming evidence from longitudinal studies. A 30-month analysis of over 26,000 Chinese secondary school students revealed a stark paradox: students who used GenAI for homework saw their assignment grades rise by 18% and their completion times fall by 30%. Yet, when tested in closed-book, proctored exams, their performance dropped by 20% within six months, with the deficit worsening to 18–24% by the study’s end.
This is not a case of cheating; it is a case of cognitive atrophy. When students use GenAI to generate introductions, summarize readings, or solve problems, they bypass the "desirable difficulties"—the slow, effortful, and often frustrating cognitive work—that is essential for building durable neural pathways. The brain learns not by receiving answers, but by struggling to produce them. By outsourcing this struggle, students are not enhancing their learning; they are degrading their ability to learn.
This dynamic is reinforced by automation bias, a well-documented psychological phenomenon where humans place undue trust in automated systems, even when the system’s output conflicts with their own knowledge or judgment. GenAI tools, which generate fluent, confident, and often correct-seeming responses, amplify this bias. Students begin to trust the AI’s output as authoritative, reducing their critical evaluation and internal verification processes.
This creates a dangerous feedback loop: the more a student uses AI, the more they rely on it, and the less they practice their own cognitive skills. Over time, their internal capacity weakens, making them even more dependent on the tool. This is the cognitive debt cycle—a term coined by Choudhuri et al. (2026)—where short-term efficiency gains are traded for long-term intellectual capability, accumulating a debt that is difficult to repay.
The most concerning finding from this study is that the highest-performing students suffered the greatest losses. Their superior ability to prompt and refine AI output allowed them to produce exceptional work, masking the fact that they were not engaging in the underlying cognitive activity. By the time they faced high-stakes assessments, they were fundamentally unprepared. This suggests that the greatest risk is not to struggling learners, but to those who are most adept at using AI as a tool—because they are the most likely to deceive themselves about their own learning.
This research, therefore, is not just about understanding AI’s impact; it is about identifying the precise conditions under which GenAI use becomes a cognitive trap, and how to design educational interventions to prevent it.
Neurophysiological Evidence: What Happens in the Brain?
The behavioral findings of cognitive offloading are now being corroborated by direct measurements of brain activity. In a groundbreaking 2025 study, Kosmyna et al. examined the neurophysiological correlates of cognitive offloading in college students writing essays with GenAI assistance.
Using functional MRI, the researchers found that when students used AI to generate text, there was a significant reduction in activation within brain networks associated with creativity, cognitive control, and self-referential thought. These are the very regions responsible for generating original ideas, monitoring one’s own thinking, and feeling a sense of ownership over one’s work.
Furthermore, students who used AI showed weaker subsequent memory recall of the content they had generated with assistance. This suggests that the act of offloading doesn’t just reduce effort; it impairs the encoding of information into long-term memory. The brain, sensing that the task was handled externally, does not invest the neural resources necessary for durable storage.
Critically, the study also found that students reported a lower sense of ownership over their AI-assisted writing. This psychological disengagement is a key indicator of the cognitive debt cycle. When students no longer feel that the work is truly theirs, they are less likely to engage with it deeply, reflect on its meaning, or strive to improve it. The AI becomes the author, and the student becomes a passive editor.
These neurophysiological findings provide a biological basis for the behavioral observations. They show that GenAI use is not merely changing how students work; it is changing how their brains work. The brain’s plasticity, its ability to adapt and rewire based on experience, means that habitual reliance on AI may be reshaping the neural architecture of learning itself. This is not a transient effect—it is a potential long-term alteration of cognitive function.
The EEF’s research must therefore incorporate neurocognitive measures where feasible, to move beyond self-reported surveys and capture the true, biological impact of GenAI on the developing adolescent brain.
The Role of Cognitive Styles: Who Is Most Vulnerable?
The impact of GenAI is not uniform across all students. The EEF’s call implicitly recognizes this by encouraging research that examines differences across student populations. Recent studies, such as the one by Choudhuri et al. (2026), demonstrate that students’ cognitive styles—their habitual ways of thinking, processing information, and engaging with tasks—profoundly influence their vulnerability to cognitive offloading.
Students with high levels of technophilic motivation (a strong attraction to technology), risk tolerance (willingness to try new tools), and computer self-efficacy (confidence in their ability to use digital tools) were found to be more prone to cognitive disengagement. This is counterintuitive: these traits are often celebrated in STEM education as indicators of competence and innovation. Yet, in the context of GenAI, they become risk factors. These students are the most skilled at using AI to generate high-quality output, and they are the most likely to trust its suggestions without critical evaluation. Their confidence in their own digital skills leads them to believe they are in control, when in fact, they are being led by the AI’s output.
Conversely, students with a strong need for understanding—an intrinsic motivation to seek conceptual coherence rather than surface-level correctness—were found to be more resistant to the negative effects. These students used GenAI as a tool for clarification or as a sounding board for their own ideas, but only after they had first attempted to think through the problem themselves.
This distinction is critical for the EEF’s research. The goal is not to ban AI, but to understand how to foster productive use. The most effective interventions will likely be those that target cognitive styles, helping students develop metacognitive awareness of their own thinking habits. For example, educators could implement a "think first, then ask AI" protocol, requiring students to document their initial thoughts before consulting the tool. This simple structure can disrupt the offloading reflex and reinforce the value of internal cognitive effort.
The EEF’s call for research that considers socioeconomic disadvantage is also vital. Students from lower-income backgrounds may have less access to high-quality AI tools or less parental support in navigating their use, potentially exacerbating existing educational inequalities. The research must therefore examine not just the cognitive impact, but the equity implications of this technological shift.
Implications for Education: Designing for Cognitive Engagement
The findings from the EEF’s research will have profound implications for educational policy and pedagogy. If GenAI use is found to systematically undermine core cognitive skills, then simply allowing unrestricted access in classrooms is not an option. Instead, education must be redesigned to actively support cognitive engagement in the age of AI.
The solution is not to eliminate AI, but to redefine its role. AI should be used as a collaborator, not a substitute. This means:
- The "Think First" Principle: Students should be required to generate their own ideas, outlines, or solutions before using AI for feedback, refinement, or clarification.
- AI as a Socratic Tutor: Instead of providing answers, AI should be prompted to ask probing questions, challenge assumptions, or request evidence for claims, forcing students to engage more deeply.
- Reflection Prompts: After using AI, students should be required to write reflections on how the AI’s output differed from their own thinking, what they learned, and what they still don’t understand.
- Scaffolded, Not Substituted: AI tools should be designed with built-in cognitive scaffolds—delays before output, prompts for self-explanation, and requirements for justification—that force the student to remain cognitively active.
Furthermore, curricula must explicitly teach AI literacy as a core cognitive skill. This includes understanding how LLMs work, recognizing their limitations and biases, and developing the metacognitive strategies to use them responsibly. Students need to learn that the goal of education is not to produce the best output, but to cultivate the best thinker.
The EEF’s £2.5 million investment is not just a research grant; it is a critical intervention to prevent a generation of students from becoming proficient prompters but incapable thinkers. The findings will shape the future of learning in the digital age, determining whether AI becomes a tool for human augmentation or a force for cognitive erosion.