Everyone says we should teach kids to think for themselves. But it’s the third hour of homework, and that stubborn equation still won’t budge. Maybe ask ChatGPT? Something like: “You’re a teacher. Explain it so a seventh grader (and their mom) can understand.”<\/p>\r\n\r\n\r\n\r\n
Is it okay to do homework with a chatbot? Won’t kids stop trying? And won’t it kill their desire to learn for good, since the solution to any problem can now be had in a couple of minutes?<\/p>\r\n\r\n\r\n\r\n
According to a Pew Research Center survey of 1,458 U.S. teens conducted in the fall of 2025, nearly two thirds of teens have used chatbots, and about three in ten use them every day. For schoolwork, about one in ten does all or most of their assignments with a chatbot’s help, another 21% do some of them, and 23% do a little. The remaining 45% haven’t used chatbots for schoolwork at all.<\/p>\r\n\r\n\r\n\r\n
\ud83e\udde0 Fact:<\/strong> 59% of teens say AI cheating is a regular thing at their school. That’s one more reason to talk about honest use at home, before the chatbot becomes the default homework partner.<\/p>\r\n\r\n\r\n\r\n Harvard physicist Gregory Kestin tested an AI tutor on students in an introductory physics course. The students took turns learning topics in a regular class and at home with a chatbot. With AI, the learning gain was more than twice as high, and it took less time: a median of 49 minutes versus about an hour in class.<\/p>\r\n\r\n\r\n\r\n \ud83e\udde0 Surprising:<\/strong> with AI, students found learning not only more useful but also more interesting. They rated their engagement 4.1 out of 5, compared with 3.6 for the in-person class. And 83% said the chatbot’s explanations were as good as a human instructor’s, or even better.<\/p>\r\n\r\n\r\n\r\n But the tutor was more than just a chat. The researchers fed it correct solutions in advance so it wouldn’t make things up, and told it to answer briefly and show one step at a time. After an explanation, it asked the student to repeat the step, and it never did the tasks for them.<\/p>\r\n\r\n\r\n\r\n A caveat:<\/strong> the participants were college students, not schoolchildren, and they studied two topics over two weeks. But the direction is telling.<\/p>\r\n\r\n\r\n\r\n Now for the opposite example. Researchers from the University of Pennsylvania, including Hamsa Bastani, ran an experiment in Turkey. Nearly a thousand high school students took part, studying math. They were split into three groups: one studied without AI, one with a regular chatbot like ChatGPT, and one with a tutor chatbot that gave hints instead of ready-made answers.<\/p>\r\n\r\n\r\n\r\n With the regular chatbot, students solved practice problems 48% better than the group without AI. But when they took a test without the chatbot, they scored 17% lower than the kids who studied the old-fashioned way.<\/p>\r\n\r\n\r\n\r\n The hint-giving tutor boosted practice results by 127%, and on the test students did about as well as those without it. So there was no harm, and there was help.<\/p>\r\n\r\n\r\n\r\n \ud83d\udca1 The “I got it” trap:<\/strong> students who used AI were overly confident that they had learned a lot. A ready-made answer creates a pleasant feeling of understanding, even though the chatbot did the thinking.<\/p>\r\n\r\n\r\n\r\n Picture two seventh graders with the same problem. The first types into the chat: “Solve it.” Ten seconds later he has a ready solution, which he copies into his notebook. The second types: “Explain where to start, but don’t solve it for me.” He tries, makes mistakes, asks again and finally solves it himself. These scenes are, of course, hypothetical, but this is exactly the difference between the two modes that both studies showed.<\/p>\r\n\r\n\r\n\r\n \ud83e\udde0 The main point:<\/strong> AI helps when it helps a child think. It harms when it thinks instead of the child.<\/p>\r\n\r\n\r\n\r\n Probably not. A chatbot doesn’t get tired, doesn’t sigh in the third hour, and is ready to explain the same thing for the tenth time. A child can also ask a “silly” question without feeling embarrassed. That’s a huge plus, if used the right way. This is our editorial observation, not a research finding.<\/p>\r\n\r\n\r\n\r\n These are our editorial tips drawn from the research, not direct conclusions by the scientists.<\/p>\r\n\r\n\r\n\r\n \ud83d\udca1 Three prompts worth copying:<\/strong> A chatbot is like a calculator: you first have to understand what you’re doing. Then it saves time. If you don’t understand, it just makes mistakes faster. The families that win are the ones that have agreed on rules.<\/p>\r\n\r\n\r\n\r\nWhere it helps<\/h2>\r\n\r\n\r\n\r\n
Where the catch is<\/h2>\r\n\r\n\r\n\r\n
The difference between a hint and an answer<\/h2>\r\n\r\n\r\n\r\n
So should we be afraid?<\/h2>\r\n\r\n\r\n\r\n
What parents can do<\/h2>\r\n\r\n\r\n\r\n
\r\n
“Explain where to start, but don’t solve it for me.”
“Give me a hint, not the answer. If I get it wrong, show me where.”
“Ask me three questions to check whether I understood the topic.”<\/p>\r\n\r\n\r\n\r\nIn place of a conclusion<\/h2>\r\n\r\n\r\n\r\n
Sources<\/h2>\r\n\r\n\r\n\r\n
\r\n