The Homework Looks Perfect. The Learning Never Happened. How to Catch an AI Faking Your Child’s Progress.

In one trial, students with unguarded ChatGPT scored 48% higher on practice and 17% worse on the exam taken alone. Check who is doing the thinking. The screen was never the variable. Tweet Now Follow Learning Success

What one study caught in the act

In 2025, researchers ran a rare direct test of AI homework help. Bastani and colleagues gave about one thousand high school students in Turkey ChatGPT for math practice. Then they measured an exam taken without it. During practice, the AI group looked far stronger, with scores 48 percent higher than the control group. On the exam, they scored 17 percent worse than students who had practiced on their own.

The same trial tested a second version, a tutor mode with guardrails that offered hints and feedback instead of finished answers. With that version, the exam harm vanished. This is one study, and one study settles nothing about every tool or every child. What it shows is how the trap looked in one real classroom: polished practice on top, borrowed thinking underneath.

There is a name for the ingredient the unguarded version stripped out. Robert Bjork spent decades showing what he calls desirable difficulties: conditions that slow visible performance today while deepening what a student keeps. Struggle is not a sign that something is wrong. For a developing brain, productive struggle is part of how the wiring changes. A tool that removes the struggle removes the learning that was hiding inside it.

Laura Lurns Learning Success Expert

Students given unguarded ChatGPT access scored 48 percent higher during practice and 17 percent worse on the exam taken without it. - Bastani et al., PNAS, 2025

Skill grows in the feedback loop

Why does one design build skill while the other borrows it? VanLehn’s review points at the answer: it is about where the feedback lands. Software that responds at the exact step where a student goes wrong comes close to the benefit of a human tutor. Software that waits until the end and marks the answer helps far less. The moment of confusion is the teachable moment, and a tool either meets your child there or misses them.

Feedback is powerful and fickle at the same time. When Wisniewski, Zierer and Hattie pooled the feedback research in 2019, the average benefit was positive but wildly uneven. An earlier review by Kluger and DeNisi found that about a third of feedback attempts made performance worse. So the useful question is never whether a tool gives feedback. It is whether the feedback responds to your child’s actual thinking, at the moment the thinking happens.

This is also where the school system leaves a vacuum. A worksheet goes in, and the response often comes back long after the thinking that produced the mistake has faded. Schools are still working out their own readiness for these tools. That slow loop is exactly the gap AI homework apps are marketed into. At home, you decide what kind of feedback your child gets tonight, and that puts you ahead of the whole debate.

Four questions that tell you who is doing the thinking

You do not need a research degree to sort a coach from a shortcut. The sorting question comes from the world of school supports: is this building the skill, or replacing the expectation that it gets built? In our view, a support that quietly removes that expectation invites dependence instead of growth. Four quick checks put the question to work while you watch your child use the tool.

• Does my child produce the answer? If the tool hands over finished work, remember the pattern from the one direct trial: practice looks better while the skill falls behind.
• Does it respond to the step where they went wrong? Step-level coaching is what let software rival human tutors in VanLehn’s review. Answer-checking alone fell far short.
• Does it feel a little hard? Bjork’s work on desirable difficulties links practice that feels slightly effortful to skill that lasts. Easy, smooth sessions fade fast.
• What does “personalization” mean here? Adjusting the level of the work and the feedback to where your child is: useful. Matching a supposed learning style: that idea was tested and did not hold up, back in 2008 and again in 2024.

That last check matters because the myth refuses to retire. Pashler’s team went looking for credible evidence that style-matching improves learning and found none, and that was 2008. Newton and Salvi reviewed surveys of more than 15,000 educators across 18 countries in 2020 and found nearly nine in ten still believe the idea works. That is not a science problem. It is a systems problem, and marketing follows belief. So bring the same healthy doubt you would bring to a brain-training app: the test of a tool was never whether it is on a screen. The test is whether your child’s own skill grows when the screen goes dark.

You want a child whose confidence rests on knowing their own mind did the work. What stands in the way is not AI. It is a framing that treats finished homework as the finish line. It is an industry that profits when convenience passes for learning, and a school system still teaching an idea that failed testing in 2008. You are the one person placed to ask the better question at the kitchen table tonight. Nobody will ever advocate for your child as hard as you will. That is true of every system, everywhere, always, and it is exactly why your involvement is not optional. If you want a partner in that work, one built on real skill-building instead of points and shortcuts, explore All Access. Start where your child actually is.

Is Your Child Struggling in School?

Get Your FREE Personalized Learning Roadmap

You describe what you see at home. We turn it into a plan you start this week.

✓ Answer 5 short questionnaires about what you already notice—30-45 minutes, at your own kitchen table

✓ Your child sits no test and gets no score—nothing to schedule, nothing for them to dread

✓ You do the answering, the AI does the writing, and a person reviews it before it reaches you

✓ Access all 40+ courses instantly—reading, math, focus, processing & more, with new ones added regularly

Why we use AI, plainly: it writes from a knowledge base our team maintains and audits—we work through it line by line and pull anything the evidence stops supporting. Doing that by hand for every family would take weeks each. This way the roadmap you get on Tuesday reflects what we corrected on Monday, and a human still reads it before you do.

Your school district must evaluate your child free of charge if you ask in writing, whatever your income and whatever the outcome (US, 34 CFR 300.111 and 300.301(b)). That route takes time and answers a different question than you do. This one starts today, from what you already know.

You get it FREE with your trial.

Get My FREE Assessment & 7-Day Trial

Your answers stay yours. We do not sell your personal information, and we do not hand identifiable assessment data to outside AI companies to train their models. See our privacy policy.

Cancel anytime during your 7-day free trial—no risk.

A screener is a starting point, not a diagnosis. If your child might need formal accommodations (an IEP or 504 plan), or you suspect a vision, hearing, or medical cause, pursue a professional evaluation too—that's the only route to those supports.