The worry is reasonable. A tool that writes the essay, solves the problem set, and summarizes the reading in seconds is an obvious temptation. But cheating is not the most important question AI raises about your child's education, and treating it as the main one leaves the deeper problem unattended. The deeper one is quieter: when AI can do so much of the academic work, is your child still doing the thinking?
There is now real evidence that this matters. A 2025 study, published after peer review, gave roughly a thousand secondary students one of three things during math practice: a standard AI chatbot, a version built to offer hints instead of answers, or no AI at all. The standard-chatbot group did markedly better during practice. Then the researchers took the AI away and tested everyone. The standard-chatbot students scored about 17 percent worse than those who had never used AI at all. They had not learned the material. They had watched the AI appear to learn it for them. And they could not tell, believing they had learned as much as their peers.
Now set that against a second fact. In a 2026 survey, 64 percent of teenagers reported using AI chatbots, but only about half of parents knew their child used these tools at all. So the cost of AI-substituted work is invisible twice over: to the student, who feels they learned what they did not, and to the parent, who often does not know the AI was in the room.
This editorial is about that hidden problem, not the obvious one. It is not a checklist of rules, and not a case for banning AI, which would be unenforceable and a mistake. It is a way of thinking about the distinction that matters most: whether your child uses AI as a thinking partner, with their own engagement as the spine, or as an answer machine that does the thinking in their place.
Parents usually come to us holding one of two worries, often both: that their child will use AI to take shortcuts and learn nothing, or that their child will be left behind without it. The worries seem to contradict each other, which is part of why parents feel stuck. They are about the same thing, the student's own thinking, and one kind of AI use resolves both at once.
Here is the pattern we see most clearly in our 1-on-1 work. Give two students the same chatbot. One asks it to explain a concept until she understands it, then writes the essay herself. The other asks it for the essay. The tool is identical, and a year later the two students are not. The difference does not show up in the assignment handed in. It shows up in what each can do when the tool is taken away. The first student has both the knowledge and the fluency. The second has neither, and a habit that gets harder to unwind the longer it goes uncorrected.
What the evidence says about how widely AI is used, what institutions are doing about it, and how reliable the tools are. Figures are drawn from research available in mid-2026, with sources listed at the end.
Two things follow. AI use is now near-universal and moving faster than school policy or parent awareness can keep up with, so the realistic goal is not to keep it from your child but to shape how they use it. And the institutions that once raced to detect and ban AI have largely retreated, which moves the responsibility for teaching good use closer to home.
Almost everything that matters about a student and AI comes down to which of these two patterns they fall into. The tool can be identical. The effect on the student is opposite. The study described earlier is, in essence, these two paths measured.
This one question does more than any rule about which tools are allowed. A student who used AI to understand a concept can talk about it intelligently afterward. A student who used it to produce an answer cannot. The point is not whether AI touched the work, but whether the student still did the thinking the work was meant to build. That is the standard universities are moving toward, the standard the IB already describes, and one a parent can apply at the kitchen table without any software at all.
It is the most common thing said to anxious parents, and there is something to it: calculators were once predicted to destroy children's arithmetic, and that panic proved overblown. But the comparison breaks at exactly the points a parent should care about, and seeing where it breaks is the clearest way to understand what is different about AI.
First, what gets automated. A calculator does the arithmetic, which in most lessons was never the point. The reasoning it served was. AI can do the reasoning. It can read the book, build the argument, and write the essay, the very things the assignment was meant to develop. When the tool absorbs the thinking, not a mechanical step beneath it, the learning it displaces is the learning itself.
Second, reliability. A calculator is always right. AI is not: in the study above, the standard chatbot gave a correct final answer only about half the time, delivering the wrong ones in the same confident voice it used for the right ones. A tool that is frequently wrong demands exactly the judgment that over-reliance on it erodes.
Third, visibility. When a student loses the ability to do long division, it shows. When a student slowly loses the ability to reason through a problem or build an argument from scratch, it does not, especially when the tool keeps making them feel capable. That silent erosion is what makes the study's central finding so important: the students did not merely learn less, they could not perceive that they had.
The resolution hides inside the analogy itself. What made calculators safe was not the tool but how schools integrated it, deliberately and with the thinking protected. That is the answer AI requires too: not prohibition, not free rein, but integration that keeps the student's own thinking at the center.
A guide that overstated the evidence would not deserve your trust. Here is what the research supports, where it is strong, and where it is still forming.
The strongest single piece of evidence is the study from the start: a randomized controlled trial, the design researchers treat as the gold standard for showing cause rather than mere correlation. It found a measurable loss for students who used a standard chatbot as an answer machine, and no such loss for those whose AI was built to make them think. It also fits a century of learning science: the effort of generating an answer yourself, instead of being handed it, is much of what makes learning stick. An AI that does the generating removes that effort.
One study, however good, is one study. This one looked at a single subject, in a single setting, over a short period, when the tools were newer than they are now. Other research pointing the same way is so far weaker: surveys that find associations rather than proof, and a widely shared brain-imaging study whose own authors caution against over-reading it. The direction of the evidence is clear enough to act on, but the science is young and moving quickly, and anyone claiming certainty in either direction is ahead of what is known.
The same evidence base contains real gains. A World Bank trial in Nigeria found gains equivalent to nearly two years of typical learning in six weeks, though the program paired the AI with teacher guidance and the researchers could not fully separate the two. A study of students revising their own essays found that AI feedback, given after a first draft, produced stronger writing. The pattern wherever AI helped is the same: the student did their own thinking and the AI supported it. That is not a different finding from the one above. It is the same finding from the other side.
What decides the outcome, then, is not the tool but whether the student's engagement is preserved or replaced. That is also why access alone does not help. The gains show up where AI comes with structure and guidance, and the child left alone with a chatbot is the one most likely to lose ground.
The questions parents bring us most often, and how we suggest thinking about each. These are starting points for a conversation, not prescriptions.
This describes roughly half of all families. Banning it tends to push the use underground rather than end it. A better first move is a conversation that assumes use instead of accusing it: ask your child to show you how they used AI on their last assignment, and what they let it do and did not. The answer tells you which path they are on. A child who used it to understand is on solid ground. A child who used it to produce the answer is worth gently redirecting, and the sooner the easier.
Many schools have not published a clear AI policy, which leaves families guessing. Two reference points are steadier than any handbook. The IB allows AI that helps a student learn, forbids AI used to pass off work as the student's own, and requires any AI contribution to assessed work to be disclosed. The College Board takes the same line for AP work: support yes, replacement no. Those standards apply to the qualifications your child is working toward, whatever the school has written down. A related caution: the AI-detection tools some schools use are unreliable, wrongly flagging honest work at meaningful rates and flagging second-language writers more often, so a detector's verdict is not proof. The answer to either worry is the same, transparent work that shows process, drafts, and reasoning.
Here the stakes are highest and the line sharpest. The Common Application, used by more than a thousand universities, treats submitting AI-generated content as application fraud, and some universities will rescind an offer over it. Yet AI is not entirely off-limits: using it to brainstorm topics or get feedback on structure sits in a different category from having it write the essay. The essay has to be the student's own, in their own voice, because voice is what admissions readers are trained to notice and what a generated essay most conspicuously lacks. The safe principle: AI may help a student think about the essay, but the writing and the substance must be theirs.
The public conversation about AI in education lives at two extremes: a cheating epidemic to be detected and punished, or a brilliant private tutor that will transform learning. Both are being sold, and neither is quite true: the detection approach has largely failed, and the tutor promise rests on claims that have not held up at scale. The honest position sits in the unoccupied middle, where AI is neither catastrophe nor cure and what decides its effect is how a particular student uses it.
Harland is unusually placed to hold that middle position, because our work is 1-on-1. We are not setting policy for a class of thirty or a campus of two thousand. We work with one student's actual assignments, watch how they use the tools, and can see the difference between a student thinking with AI and one hiding behind it. The answer-machine and thinking-partner patterns are not abstractions to us. They are what we watch for in a student's writing and process, week to week.
We have also tried to be honest about what is and is not known, and to say where the evidence is still thin. If your child's experience or your own reading points somewhere different, we would want to hear it. This is a fast-moving subject, and we expect to revise this guide as the evidence develops.
Harland's AI for Education program exists for exactly the problem this guide describes. It coaches students, in their own academic work, to use AI in the way that builds thinking instead of replacing it, with ethics and integrity at the foundation rather than bolted on as a warning.
Whether the worry is shortcuts or falling behind, the question underneath is the same, and it is one we work on with students every week. A short conversation can help you understand where your child is and what support would look like.
Start the conversation