How Does AI Affect Learning? When Support Becomes a Shortcut

Calculators changed math class. Search engines changed research. AI goes a step further: it can produce the reasoning and the writing itself.

That is hard to resist when a student faces a difficult assignment on a Tuesday night. Within seconds, the homework can be finished and nobody is still at the kitchen table at 11 p.m. But finishing the work and learning how to do it are not the same thing.

The difference appears later, after the tool is closed and the student has to use that knowledge again. Much of the research on AI and learning sits in that gap.

It Matters How Students Use AI

Asking whether a student “uses AI” does not tell us much. Having a chatbot write an essay is very different from asking it to identify a weak paragraph in the student’s own draft.

AI can offer a hint, explain an idea, generate practice questions or flag an error. It can also take over the reasoning and return a finished answer.

The useful question is: Does the tool help the student think, or does it do the thinking for them?

Why Learning Takes Effort

Reading the right answer can feel productive, but learning requires us to connect information, recall it, correct mistakes and use the idea again.

Researchers sometimes call this productive struggle. A meta-analysis of 53 studies found that students who attempted a problem before receiving instruction often developed a deeper understanding than those who received the instruction first.

The task must be challenging but still within reach, followed by helpful feedback. Most of this evidence comes from math and science. Frustration alone does not teach anything.

Good support works like scaffolding: it removes confusion that blocks progress while leaving the student to build the understanding. AI can provide that support, as long as it does not take over the task.

What Is Cognitive Offloading?

Cognitive offloading means using an outside tool to handle mental work, as we do with calendars, calculators and GPS. It can free up attention for more important decisions.

What matters is which work we hand over.

• Supportive offloading lets the tool handle lower-level tasks while the person continues to reason and decide.

• Dependent offloading lets the tool work through the reasoning and produce the conclusion.

A July 2026 study in Frontiers in Psychology made a similar distinction. Researchers surveyed 589 university students and early-career workers three times. Dependent use was associated with lower motivation and a greater tendency to give AI cognitive authority. Using AI to prompt, test or extend one’s own thinking was associated with better outcomes.

The study was correlational and relied partly on self-reports, so it cannot prove AI caused those differences. Both styles produced similar immediate benefits, making the difference hard to notice in the moment.

How AI Can Get in the Way of Learning

Problems are more likely when AI steps in before the student has engaged with the material.

The Student Skips the First Attempt

A first attempt shows students what they understand and where they need help. An immediate answer removes that useful information.

Familiarity Feels Like Understanding

A clear explanation can create a quick feeling of recognition: That makes sense. I know this now. A student can test that feeling by closing the tool and explaining the idea or solving a related problem.

There Is Less Retrieval Practice

Memory gets stronger when we practise retrieving information. If AI always produces the answer, the student gets little practice recalling what they know.

Confusion Becomes Harder to Tolerate

Learning often feels messy. Because AI can remove that discomfort in seconds, students may have fewer chances to build persistence and confidence with uncertainty.

Current research does not show that AI permanently damages attention spans or “rewires” the brain. Those claims go further than the evidence allows.

What the Research Shows So Far

Two studies show why simple conclusions about AI and learning are difficult.

In a field experiment published in PNAS, nearly 1,000 high school math students practised with GPT-4. They performed better while using it. Once it was removed, those who used the unrestricted version did worse than classmates who practised without AI. A tutoring version that offered guidance avoided much of that decline.

A randomized controlled trial with 194 university physics students found a different result. A custom AI tutor produced greater learning gains in less time than an active-learning class. This was not a general chatbot. It used expert materials, structured sequencing, targeted feedback and active participation.

Together, the studies suggest that AI can help when it keeps students involved and becomes riskier when it removes work they need to practice.

Long-term research remains limited, with outcomes varying by student, subject and task.

When AI Can Help

AI is most useful when it acts like a tutor: explaining an idea, asking a guiding question, commenting on the student’s answer, creating practice problems or pointing out a missing step.

For students with learning disabilities, ADHD, language barriers or working-memory difficulties, those uses can provide meaningful access. The key is whether AI removes a barrier or removes the learning.

A Five-Step Approach to Studying With AI

1. Try before asking for help. Work alone long enough to see what you know and where you are stuck. A few minutes may be enough.

2. Ask for a clue, not the answer. Request a hint, a question or feedback on your reasoning. Tell the tool you want to work it out yourself.

3. Question the response. Check it against reliable sources. Look for missing information, weak evidence and hidden assumptions. Deciding what to trust and why exercises critical thinking.

4. Put the tool away. Explain the idea or solve the problem without help. If you cannot, keep practicing.

5. Apply the idea somewhere new. Ask AI for a similar problem, then solve it before looking at the answer.

Prompts That Leave the Thinking to the Student

After making a first attempt, students can try:

• “Ask questions that help me reach the answer. Do not tell me unless I ask.”

• “Review my reasoning and identify only the first weak step.”

• “Quiz me one question at a time without showing the answer.”

• “Give me a similar problem to solve independently.”

• “Let me explain this in my own words, then identify what I missed.”

• “Give me two explanations and ask me to compare their strengths.”

Why Some Students Rely on AI More Heavily

Heavy AI use is not always about convenience. Anxiety, perfectionism, exhaustion, ADHD, executive-function difficulties and low confidence can all make it hard to begin. After years of struggling at school, avoidance may feel safer than getting it wrong again.

Dr. Phoenix Brill, MC, R. Psych., PsyD, explains how this can affect anxious and perfectionistic students:

“For a perfectionistic student, the discomfort of a rough first draft can feel intolerable. AI offers an escape from that discomfort. It hands them something polished before they’ve had to sit in the uncertainty of not-knowing. The problem is that the struggle they’re avoiding is exactly where the learning happens. Over time, leaning on AI to skip that stage doesn’t just weaken the skill; it reinforces the belief that their own unfinished thinking isn’t good enough.”

This is why “just try harder” often fails. Difficulty focusing is often tied to stress, fatigue and overload, which can make an instant answer especially tempting.

Parents and teachers should ask what purpose AI serves. Is it organizing thoughts, calming panic, supporting accessibility or completing work the student does not understand? The goal is conversation, not diagnosis.

How Parents Can Talk to Teens About AI

A blanket ban can push AI use out of sight, while unrestricted access can replace needed practice. Clear expectations and curiosity are usually more useful.

Dr. Brill explains:

“Shame can make kids defensive, and defensive kids stop talking. I’d steer parents away from ‘Did you cheat?’ and toward curiosity: ‘How are you using it? What’s it helping with, and what’s it doing for you?’ The goal is to make AI a normal thing you can discuss openly, so that when a student is genuinely stuck, coming to you feels safer than hiding behind a chatbot.”

Parents can ask:

• What part of the assignment did you do yourself?

• What did the AI help you understand?

• Could you explain the answer without opening it again?

• Where did you get stuck?

• What does your teacher allow?

• How did you check whether the information was accurate?

Distinguish between using AI to understand, practise, edit, generate work and submit generated work as one’s own. Students should follow their school’s rules. For more guidance, read about supporting family mental health.

When AI Use Points to a Larger Problem

Sometimes AI makes an existing difficulty easier to see, including anxiety, depression, ADHD, a learning difference, perfectionism, stress or burnout. Frequent AI use alone is not evidence of a mental health condition.

Dr. Brill encourages parents and teachers to look past the tool and ask what it is doing for the student. AI use becomes more concerning when it is paired with avoidance: a student who has stopped attempting hard work without it, panics when it is unavailable, or hands in work that no longer matches what they can do on their own.

“Those aren’t laziness. They’re often signs of underlying anxiety, attention difficulties, or a gap in skills the student is quietly trying to hide.”

Professional support may be worth considering when difficulties persist or affect sleep, relationships, mood, attendance or daily life. Flourish offers therapy for children and teens in Calgary. Adult learners can explore individual counselling in Calgary.

Keep the Student in the Work

AI will remain part of daily life. There is no value in making learning difficult for its own sake, but students need chances to remember information, make judgments and work through problems.

For the student at the kitchen table, the assignment may get finished either way. What matters is whether they attempted it, got something wrong and worked toward an answer.

AI can help a student reach that moment or avoid it. The difference often comes down to how it is used.

Frequently Asked Questions

Does AI make it harder to learn?

It can. If AI supplies an answer before a student has thought about the problem, it may replace the practice learning requires. When it offers structured tutoring, feedback or well-timed questions, it can help. Research so far suggests that the design of the tool and the way a student uses it matter more overall than the simple presence of AI.

Can AI improve critical thinking?

AI can help a student practise critical thinking by presenting counterarguments, questioning an assumption or identifying a weak step in their reasoning. The benefit comes from what the student does next. They need to check the tool’s claims, compare sources, weigh the evidence and decide which conclusion is best supported. Accepting a polished answer without examining it does not build the same skill.

How should students use ChatGPT for studying?

Students should make a first attempt before opening ChatGPT. Once they know where they are stuck, they can ask for a hint, a question or feedback rather than a complete answer. Important claims should be checked against class materials or reliable sources. Finally, the student should close the tool and explain the idea or solve a similar problem without help.

Should parents ban AI for homework?

Rules should reflect the student’s age and needs, the purpose of the assignment and the school’s policy. A total ban may push AI use out of sight without teaching responsible habits. Unrestricted access can replace practice a student needs. Clear limits, such as requiring a first attempt and discussing how the tool was used, are often more useful than either extreme.

If school-related stress, avoidance or anxiety is becoming a pattern in your family, get in touch with Flourish to discuss support.

This article is for general education and does not replace an assessment or advice from a qualified professional.

Research referenced

• Bastani, H. et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26).

• Kestin, G. et al. (2025). AI tutoring outperforms in-class active learning. Scientific Reports, 15.

• Zhu, Q. et al. (2026). Not all cognitive offloading is equal: distinguishing dependent and autonomous offloading to generative AI. Frontiers in Psychology.

• Sinha, T. & Kapur, M. (2021). When problem solving followed by instruction works: Evidence for productive failure. Review of Educational Research, 91(5).