Helping parents navigate their children’s academic development in the age of AI

Introduction

Recent data reveals that global attainment levels in the core academic disciplines of literacy, numeracy and science have had a marked decline. Many have speculated that AI could be to blame. This piece will address the current state of research about the degree to which AI educational tools are negatively impacting student attainment and what parents can do to offset these impacts.

This article will also draw on my own experiences as a tutor  and my many interactions with parents who are concerned about their children’s academic performance. I receive questions regularly on how parents should introduce AI to their children, what limits may be needed and how to best adapt their education as the technology changes. There are continuous pressures on parents to find a technological edge for their children’s education. However, it is important to not treat AI as a tool which has authority over  parents, educators and students. AI is a tool that should be integrated within a system of educator-led and data-backed approaches to learning. A key takeaway will be that learning the ‘hard way’ is even more necessary as the role of ed-tech in education continues to expand.

The current academic backdrop

48% of students in OECD countries use AI chatbots weekly to help them learn.[1] Simultaneously, PISA, the OECD’s Programme for International Student Assessment, recorded in their 2025 assessment the lowest OECD-average scores so far in science, reading and maths. [2] We should not be quick to draw a causal link between these two facts, but it does warrant asking whether the increased use of AI chatbots may be contributing to falling average scores. What the evidence does show is that students are spending less time on, and retaining less of, the tasks that AI can most easily solve, such as text-based maths problems.[3]

How is it possible that students, with easy access to inexpensive AI chatbots and other educational tools that seemingly offer unlimited and personalised learning support, are experiencing significant drops in foundational examination scores? Technology has enabled a revolution in academia around the world and AI is the most capable iteration of that technology. It is not so farfetched to believe how this revolutionary technology would result in stronger grades round the board.

The data is starting to show a more complex picture on AI in education than previously understood. There is evidence that suggests adverse impacts on student performance from AI use. Students who use AI for specific schoolwork tasks, such as summarising texts, drafting or conducting research, “score around 20 points lower in science, on average, than those who do not.”[4] Shockingly, students who don’t use AI are equivalent to being around ‘one year of schooling ahead of their peers’.[5] Reading comprehension, a skill that matters arguably the most in the AI age, has declined. This will have secondary impacts on other areas such as evaluating information, connecting multiple sources and thinking critically about texts. Limited or moderate use of devices for learning is often linked to better outcomes, while leisurely use at school links to substantially lower performance. The OECD stresses that these are associations, not proof that AI causes lower scores. However, this underscores that how a child uses AI matters.

Hyper-convenience and reduced persistence in students

Emerging data suggest that unrestricted use of technology can undermine educational objectives. The first problem stems from the extreme convenience of educational technology, causing students to persist less in their learning. It is plausible that immediate access to AI chatbots for a student’s problem solving can hinder their educational performance. This relationship seems to strengthen as students gain more independent access to AI. One large study found the biggest effect among high school and university students, and none among the youngest pupils.[6]

Students were shown to spend substantially less time on problems which AI could solve easily, which was associated with a “substantial decline in retention of those underlying concepts.”[7] So, it is equally important that students are taught to see the value in spending enough time on these subject areas. In a deeper sense, there is a value problem that needs addressing. In the same way a child in Key Stage 1 needs to avoid a calculator when they are adding basic numbers, an early high school student will need to avoid AI to learn the capability themselves. If access to AI tools remains unrestricted, students may see little point in learning these concepts, methods and faculties themselves. This will only result in further declines in those subject areas.

Randomised trials[8] have provided causal evidence for two key consequences of the use of AI: reduced persistence and impairment of unassisted performance. Whilst AI assistance improves performance in the short term in areas such as mathematical reasoning and reading comprehension, once it is removed, people perform significantly worse and are more likely to give up on problems after as little as ten minutes of use. This finding matters because effort and persistence are two of the strongest predictors of long-term learning, both foundational to more complex skill development. There is growing literature on this phenomenon, termed cognitive offloading, which may be gradually leading to the deskilling of students. Long-term skill, presented through effort and persistence, needs to be prioritised.

The impact of this cognitive offloading is demonstrated in various experiments in both maths and reading where, once the AI was removed, “solve rates dropped and skip rates increased across all subjects.”[9] It is very easy for students to develop a reliance on these tools, where higher skip rates in tasks, particularly in areas such as reading comprehension, ‘suggests that AI assistance reduces persistence and motivation’.[10] Earlier randomised trial evidence points the same way: Bastani et al. (2025) found that unrestricted GPT-4 access during maths practice improved high school students’ performance while they had it, but reduced their unassisted exam scores by 17%.[11] Students who solved maths tasks without assistance showed greater persistence.

The term cognitive surrender helps us understand this phenomenon, where unlike a tool like a calculator which delegates a discrete subtask to a tool, an AI tool takes over “the cognitive control of problem-solving.”[12] Where a calculator or dictionary outsources a narrow task, the student “retains the structure of reasoning.”[13] They are still in control of the solution strategy, evaluate the results and then, most importantly, “integrate it with prior knowledge”[14] to produce the learning effect. With AI, by contrast, the student “relinquishes the cognitive control of problem-solving.”[15] As the researchers put it, “A student who copies a word problem into a chatbot and pastes the returned solution to the learning system bypasses required cognitive processing, intermediate evaluation, and the retrieval practice that converts exposure into memory.”[16]

There is a secondary impact from AI assistance, as it can “impair unassisted performance for a task that is closely intertwined with critical thinking”.[17] If persistence isn’t practiced, the need for critical thinking will reduce. Critical thinking requires precise and coordinated thought processes, and take time and effort to develop.

AI access also reshapes how students approach learning a new topic. In one experiment, students spent “less time drafting text and more time reading and searching for information and report higher enjoyment.”[18] AI access is fundamentally reshaping how students how much importance students attach to learning concepts. In fact, an interesting mechanism has been observed. When AI completes a task almost instantaneously, the student who observes this will naturally alter their reference point for how long a task should take.[19] A shift takes place where unaided work starts to feel more effortful. This becomes a self-enforcing mechanism: “each act of offloading shifts the reference point, increases the subjective cost of unaided effort, and makes future offloading more attractive.”[20] This demonstrates a compounding effect of continued AI use for task completion. The tolerance for effort diminishes and the space for effort reduces. Students become less willing to struggle with hard problems.

The critical point from the studies I have referenced, as well as other mounting evidence, is that assistive educational tools are becoming normalised in private and school settings. These tools credit student performance based on results that may be driven by AI inputs. When a platform sees a student performing answers well and quickly, it does not see that they may be using AI to bypass the learning process, skewing their performance and compounding learning issues down the line. The implication of this is a “structural disruption of the adaptive logic that makes intelligent tutoring systems effective.”[21] The platform’s measurement of student attainment becomes vulnerable to instances where students do not actually possess the knowledge. This is the age where educational tech tools can can categorise student targets in an almost infinite set of metrics, but there is a real conundrum here: can we trust the same systems that are facilitating students’ cognitive offloading to assess their performance? Going back, if we now apply this to the PISA score averages from 2025, one possible explanation here is the potential dependency on AI tools may have inflated their performance in homework and assisted tasks. When it came to unassisted examinations, the drop off in performance may highlight that they did not sufficiently learn the processes needed in the first place.

The Abundance of educational tech tools

Parents often ask what tools can provide an academic edge. The abundance of these available products has generated an unnatural competitiveness that didn’t exist a few years ago. I am constantly surprised by the introduction of new products and services, and there seems to be no slowdown. Parents are often surprised when I suggest that, based on the evidence, regulating AI’s usage will best serve their children’s interests.

I’m finding that my role as tutor is in part mitigating the damaging effects of these technological tools. Students are surrounded by an unregulated flurry of tech resources, which is creating dependence on the tools and less motivation to do the hard work. The main driver of these resources is the claim I often read on company websites that they can deliver results to students faster than the traditional methods of teaching. Research shows that intelligent tutoring systems and adaptive learning platforms can produce moderate-to-large gains when students engage with them authentically. However, generative AI programmes can be “used to produce correct answers without the underlying cognitive engagement.”[22] In other words, the programme’s measure of success of student performance is based on its own inherent calculation of subject mastery, which can become misleading and miss the other key learning criteria needed for student performance.

A more obscure issue is AI’s covert integration into technological devices, where AI features work in the background and offer answers before a student has even asked for assistance. I recently experienced a perfect example of this in one of my own lessons. I was teaching a maths concept on my iPad and noticed that as I was writing the problem, it immediately prompted if I wanted the answer. My student, almost subconsciously, clicked the prompt, magically revealing the answer. Even the function of pulling out a calculator, manually pressing the buttons and then writing the answer seem to cause too much inconvenience. How can I possibly blame the student for this behaviour when the solutions are right there, all the time? This can feed a habit where students can very easily lose control over their critical thinking faculties.

The iPad example highlights an inherent tension between technologies that offer instant answers and the core mechanisms of learning. As students spend less time pondering the problems and their solutions, they will naturally spend less time thinking about the subject itself. Learning, at times, can be a messy, non-linear process where wrong attempts can ignite the correct solutions. Curiosity and intrigue emerge when you really let the ideas marinate in your brain. Getting lost in methods, trial and error, questioning bad logic and going on tangents are all valuable and essential processes that are fundamental to the learning experience.

The less time we think about problems within fields of study, the less time we commit to building our academic interests. So naturally, student attainment is a function of time spent in that subject, and this may partially explain the recent drops in global scores. By outsourcing the thinking part to an AI agent, students lose access to all those unquantifiable and messy processes which are critical to their cognitive development.

Of course, there are other contributing factors including rising mental health issues, record SEND diagnoses, class sizes, teaching quality, economic resources and widening inequalities across societal groups. However, many of these contributing factors are to a large degree being compounded by the constant availability of technology. The more children offload cognitive processes to AI, the less they hone their own ability to think and solve complex problems unaided.

The role of educators

When students are asked in observational studies about their use of AI, in general they report that AI improves their learning. Yet these beliefs may not track actual effects[23] where even experienced professionals can misjudge AI’s productivity effects. As a tutor, it is important to know when not to help. Part of our job as educators is to restrict access to loopholes that get in the way of learning. There is a constant balancing act of support and resistance, that depends on the student’s capabilities and challenges. Some students are very independent and need less support and some thrive with more of a supported presence. It is important to identify these nuances between students and plan lessons accordingly. An effective mentor will encourage development by adjusting the type and frequency of help given[24]. This should now extend to restricting use of AI if this is potentially compounding the student’s academic challenges.

What I recommend parents when it comes to their children’s learning

1. Keep prioritising all foundational areas of study

As attainment levels are dropping, it is crucial to maintain basic levels of numeracy, literacy and written comprehension. These are the core tools which provide the basis for more specialist skills to emerge. As discussed, prioritise subjects with equal importance, regardless of how effective AI is in them.

2. Valuing curiosity even more

Get your children to explore ideas and subject areas. Spark interest and get children to think deeper about areas they find interesting. See if they can connect their areas of curiosity together. Education doesn’t exist in a vacuum, and the relativity of ideas is how a child can develop more wide-ranging curiosity in different academic fields. Help them find comfort in riskier ideas.

3. Value independence of thought

Make sure that there are opportunities to work independently. Students won’t learn what they are capable of, and the degree they can show persistence, if there is AI at their disposal.

4. Be deliberate in how you use AI in your children’s learning

Working in the education sector, I am increasingly bombarded with the messaging that students should learn how to use AI as they are entering a tech-oriented world. I have read countless articles on how the need to familiarise children with tech is crucial, otherwise they will be left behind. However, I firmly stand on the belief that this should be a secondary, not a primary, goal.

Be stringent on what it should not be used for. Learning how to use tech is crucial, but its very success is based on the academic confidence of the student utilising it. Academic agency must come first, as this will lead to comfort with being in control of ideas, argumentation and probing for solutions. If those skills aren’t adequately taught first, there may be a risk of tech overdependence, and the student ultimately stops thinking like an academic and starts acting as a responder to tech agents. When this foundation is there, as has been the primary goal of education up till now, the power of tech can be better harnessed.

Learning how to use tech is meant to be intuitive. These tools are designed to be easy to use and apply to daily needs. The things that need to be learned are resilience, research, evidence, seeing if a source is credible, being organised, being curious, asking questions and doing homework.

While the capacities of these tools are expanding, it is better to restrict its use for areas where they can provide a concrete advantage. For example, if a student has tried very hard but doesn’t understand a tricky concept, I would be more apprehensive about getting a chatbot to explain it and would always defer to teachers or tutors, who can find exactly what is causing the conceptual gap. However, if a student simply wants more practice questions in an area they understand, AI is great at setting question types, instead of endlessly scrolling Google for papers. By thinking of AI in this way, parents and students gain control and reassurance that they are using the technology in a way that is positively impacting their learning.

A more scalable approach is to design learning tasks deliberately.[25] If work is designed so that it is inherently resistant to AI assistance, it can authentically preserve engagement without requiring supervision.[26] This is where support may be needed from tutors and teachers to make sure the learning processes are authentic and leading to real progress.

5. Build resilience by thinking less on outcomes and instead on values, skills and processes.

The issue with many tech-based learning platforms is the fixation on the outcomes of learning. I discussed how this can skew metrics on student performance if they are overly relying on AI in the first place. One way to counter this is to separate processes from the outcomes of learning. Are students able to make arguments, apply logical thinking to problems, comprehend a piece of literature and have a discussion, work on a maths solution using various methods and so on and so forth? These skills take the onus away from the outcome and start to build a holistic system that incorporates various skills and processes within their general learning. This is a very effective way to interrupt the constant search for perfect answers and scores, which my own students in high school have responded positively to. This also instils a sense of agency over one’s own education, as these processes require deliberation and an awareness of how everything connects.

6. Follow the first attempt rule.

Do things on the first attempt yourself with no assistance. Personal statements, essay plans, solution probing, citations, grammar and spelling checks. This habit should be extended to the most mundane tasks, especially for younger students who are learning these skills for the first time. This helps resist the temptation of using an AI agent and hardwires the sense of control over one’s academic goals.

7. Socialise learning for your children

Learning is both an independent and a social experience. One way to get children to take agency over their education is by getting them to work with other peers, for example on homework, presentations or a group project. Public libraries are great places for this. When children think aloud and debate with peers, they're doing the opposite of outsourcing their thinking. John Dewey, in Democracy and Education (1916), puts it aptly:

 "Not only does social life demand teaching and learning for its own permanence, but the very process of living together educates. It enlarges and enlightens experience; it stimulates and enriches imagination"

9. Ask your child’s school about its AI policy

Ask schools directly what their policy is on AI usage. Schools are authorities and their policy measures will make it easier for parents to apply the rules at home too. 

10. Stay up to date on new research on AI in education

This is a fast-growing area of research, with new programmes constantly entering classrooms and homes, and new studies regularly changing our understanding of how AI affects learning. What we know today will certainly look very different in a year's time. As a tutor, I make a point of following this research closely, and I'll be keeping you informed through my blog, sharing the findings that matter most for your children and what they mean in practice. If you'd like to stay updated, subscribe to my newsletter or follow me on LinkedIn.

In conclusion

While attainment levels are seemingly dropping, it is critical that we don’t rush to conclusions about the causes. However, we should collectively take more responsibility, recognising that the solutions are in the hands of educators and parents more than ever.

There is growing literature on the misalignment in objectives between human preferences and AI-dependent systems. Sustained AI use risks eroding the very essential capacities needed for a well-rounded student and, more profoundly, displacing human participation over time.[27] It is important to have conceptual mastery of the foundational academic skills as a developmental prerequisite for the other, more refined and complex processes. Without basic skills, such as comprehension and numeracy, built on the foundation of persistence, effort and responsibility, higher-order competencies such as critical reasoning, algebra and more remain inaccessible.[28]

Learning is a set of building blocks, each resting on the one before, and in an age where AI is programmed to skip these steps almost instantaneously, the processes of learning should be valued even more highly.

References

Bastani, H., O. Bastani, A. Sungu, H. Ge, Ö. Kabakcı, and R. Mariman. "Generative AI without Guardrails Can Harm Learning: Evidence from High School Mathematics." Proceedings of the National Academy of Sciences 122, no. 26 (2025): e2422633122. https://doi.org/10.1073/pnas.2422633122.

Contractor, Z., and G. Reyes. "Experimental Evidence on the Learning Impact of Generative AI." Preprint, arXiv, 2026. https://doi.org/10.48550/arXiv.2607.08849.

Dewey, John. Democracy and Education: An Introduction to the Philosophy of Education. New York: Macmillan, 1916.

Liu, G., B. Christian, T. Dumbalska, M. A. Bakker, and R. Dubey. "AI Assistance Reduces Persistence and Hurts Independent Performance." Preprint, arXiv, 2026. https://doi.org/10.48550/arXiv.2604.04721.

OECD. PISA 2025 Results (Volume I): Future-Ready Students. Paris: OECD Publishing, 2026. https://doi.org/10.1787/73451bc5-en.

Rismanchian, S., H. Uzun, J. Matayoshi, E. Cosyn, and E. Kurd-Misto. "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build." Preprint, arXiv, 2026. https://doi.org/10.48550/arXiv.2605.21629.

[1] OECD, PISA 2025 Results (Volume I): Future-Ready Students (Paris: OECD Publishing, 2026), https://doi.org/10.1787/73451bc5-en.

[2] OECD, PISA 2025 Results.

[3] S. Rismanchian et al., "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build" (preprint, arXiv, 2026), [page],https://doi.org/10.48550/arXiv.2605.21629.

[4] OECD, PISA 2025 Results (Volume I): Future-Ready Students (Paris: OECD Publishing, 2026), https://doi.org/10.1787/73451bc5-en.

[5] OECD, PISA 2025 Results.

[6] S. Rismanchian et al., "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build" (preprint, arXiv, 2026).

[7] Rismanchian et al., "Faster Completion, Less Learning," 13.

[8] G. Liu et al., "AI Assistance Reduces Persistence and Hurts Independent Performance" (preprint, arXiv, 2026), https://doi.org/10.48550/arXiv.2604.04721.

[9] Liu et al., "AI Assistance Reduces Persistence," 5.

[10] Liu et al., "AI Assistance Reduces Persistence.”

[11] Liu et al., "AI Assistance Reduces Persistence.”

[12] S. Rismanchian et al., "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build" (preprint, arXiv, 2026), 11.

[13] Rismanchian et al., "Faster Completion, Less Learning," 11.

[14] Rismanchian et al., "Faster Completion, Less Learning," 12.

[15] Rismanchian et al., "Faster Completion, Less Learning."

[16] Rismanchian et al., "Faster Completion, Less Learning," 12.

[17] G. Liu et al., "AI Assistance Reduces Persistence and Hurts Independent Performance" (preprint, arXiv, 2026), 8, https://doi.org/10.48550/arXiv.2604.04721.

[18] Z. Contractor and G. Reyes, "Experimental Evidence on the Learning Impact of Generative AI" (preprint, arXiv, 2026), 25, https://doi.org/10.48550/arXiv.2607.08849.

[19] G. Liu et al., "AI Assistance Reduces Persistence and Hurts Independent Performance" (preprint, arXiv, 2026), https://doi.org/10.48550/arXiv.2604.04721.

[20] Liu et al., "AI Assistance Reduces Persistence," 10.

[21] S. Rismanchian et al., "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build" (preprint, arXiv, 2026), 13.

[22] Rismanchian et al., "Faster Completion, Less Learning," 11.

[23] Z. Contractor and G. Reyes, "Experimental Evidence on the Learning Impact of Generative AI" (preprint, arXiv, 2026), [page], https://doi.org/10.48550/arXiv.2607.08849.

[24] G. Liu et al., "AI Assistance Reduces Persistence and Hurts Independent Performance" (preprint, arXiv, 2026), [page], https://doi.org/10.48550/arXiv.2604.04721.

[25] S. Rismanchian et al., "Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build" (preprint, arXiv, 2026).

[26] Rismanchian et al., "Faster Completion, Less Learning."

[27] G. Liu et al., "AI Assistance Reduces Persistence and Hurts Independent Performance" (preprint, arXiv, 2026), https://doi.org/10.48550/arXiv.2604.04721.

[28] Liu et al., "AI Assistance Reduces Persistence."