Dog ate my homework

Some 54% of American teenagers have used a chatbot for schoolwork, and about one in ten do all or most of it that way. It is rare for a technology to move from novelty to homework habit in three years. The figures come from a Pew Research Center survey of 1,458 American teenagers, fielded in September and October 2025.^{2Pew Research CenterHow Teens Use and View AI24 Feb 2026} A year earlier, 26% had used ChatGPT for schoolwork; the year before that, 13%.^{1Pew Research CenterAbout a quarter of U.S. teens have used ChatGPT for schoolwork, double the share in 202315 Jan 2025} The measures differ (one chatbot, then any), so the line on our chart is not strictly a line. Its direction is not in doubt.

We think the right response is neither the panic nor the shrug. The panic is wrong because the thing it fears has not obviously happened. The shrug is wrong because something else has.

Homework, assisted. American teens using AI for schoolwork, %. 2023 and 2024: ChatGPT; 2025: any chatbot

	Teens using it for schoolwork
2023	13
2024	26
2025	54

Source: Pew Research Center¹²

Take the panic first. Pew found that 59% of teenagers say cheating with AI happens at least somewhat often at their school, and 76% of those who have used chatbots for schoolwork say their classmates cheat with them at least sometimes.^{2Pew Research CenterHow Teens Use and View AI24 Feb 2026} That sounds damning until one recalls that teenagers cheated before. A Stanford survey of students at 40 American high schools found that about 60% to 70% of students had engaged in cheating behaviour in the last month, a rate the report called the same or slightly lower since ChatGPT’s debut. “We are seeing little evidence that the needle has moved for high schoolers overall,” said Victor Lee, Stanford’s faculty lead on AI and education.^{4CNN, via ABC7Controversial AI ChatGPT didn't increase cheating in high schools, Stanford study finds13 Dec 2023} The data are from 2023, early in the tool’s life, and should be repeated. But the burden of proof lies with those claiming a surge.

Teenagers also draw lines. In 2024, 54% called ChatGPT acceptable for researching a new topic, 29% for solving maths problems and 18% for writing essays; 42% thought essays unacceptable.^{1Pew Research CenterAbout a quarter of U.S. teens have used ChatGPT for schoolwork, double the share in 202315 Jan 2025} That is a hierarchy of sin that any housemaster would recognise. It does not prove virtue, but nor does it describe a generation with no scruples.

Now the shrug. The use is not evenly spread. About six in ten Black and Hispanic teenagers have used chatbots for schoolwork, against about half of white teenagers.^{3Pew Research CenterDemographic differences in how teens use and view AI24 Feb 2026} One in five teenagers in households earning under $30,000 do all or most of their schoolwork with a chatbot, against 7% in the better-off households.^{3Pew Research CenterDemographic differences in how teens use and view AI24 Feb 2026} The same tool is a tutor for one child and a crutch for another. A family that can pay for a human tutor has a choice; a family that cannot has a chatbot.

This is the point that school policies mostly miss. Banning the tool from school networks does nothing about the phone in the pocket. Detection software offers false comfort. What remains is to change the assignment. Work done in class, on paper, in front of a teacher, tests whether a pupil can think. Work done at home tests whether a pupil can prompt. Both are skills, but only one is on the exam.

The best objection to all this is that learning requires struggle, and a tool that removes the struggle removes the learning. That is plausible, and nothing in these surveys refutes it, because none of them measures learning. Pew asked teenagers how helpful the tools were; about a quarter said extremely or very.^{2Pew Research CenterHow Teens Use and View AI24 Feb 2026} Pupils are poor judges of what they have not yet learnt. Schools should measure outcomes, not usage, and until they do, the cheating statistics tell us more about adult anxiety than adolescent behaviour.

The dog, in the end, was always going to eat the homework. The sensible teacher sets work the dog cannot reach.