Surveys show the main classroom worry about AI is already common practice: a Pew Research Center poll taken in autumn 2025 found that 54 percent of American teenagers had turned to chatbots for homework or classwork, and one in ten relied on them for nearly all of it. Whether that use harms learning is less settled. The most cited brain-activity study of essay writing with a chatbot is an unreviewed preprint with 54 participants.
Key takeaways
- Pew's February 24, 2026 report, drawn from 1,458 teens, put chatbot use for schoolwork at 54 percent, with 21 percent using them for some of it and 10 percent for all or most.
- Fifty-nine percent of teens believe classmates cheat with AI at least somewhat often, and the figure reaches 76 percent among teens who themselves use chatbots for schoolwork.
- A 2025 MIT Media Lab preprint reported weaker brain connectivity and poorer recall of their own essays among chatbot users, in a sample of 54 participants.
- Self-reports and small preprints show what is happening and what to test, not what AI does to learning in general.
- No official Canadian figure on student use was found, so Canadian schools cannot simply borrow the U.S. numbers.
What students say they do
The Pew Research Center published "How Teens Use and View AI" on February 24, 2026. Ipsos ran the online survey in late September and early October 2025 among 1,458 teens and their parents, with a margin of error of 3.3 percentage points. Of the teens, 45 percent had never used chatbots for schoolwork, 23 percent used them a little, 21 percent for some tasks, and 10 percent for all or most. About four in ten or more used them to research a topic or solve math problems, and 35 percent used them to edit writing they had done.
The table lays out the numbers in one place.
| Teens in the Pew sample (fall 2025) | Use of chatbots for schoolwork |
|---|---|
| 45 percent | None |
| 23 percent | A little of their schoolwork |
| 21 percent | Some of their schoolwork |
| 10 percent | All or most of their schoolwork |
Cheating and perceived fairness
The same survey asked about cheating. Fifty-nine percent of teens thought that peers at their school cheat with AI at least somewhat often, and 76 percent of teens who had used chatbots for schoolwork agreed. These are perceptions, not audits of actual misconduct. They still matter for schools, because a classroom where students believe that others cheat with AI changes how fair grading feels. About a quarter of all teens called chatbots extremely or very helpful for schoolwork, and another quarter called them somewhat helpful, so the same tools are seen as useful and as a risk.
One study of what happens in the brain
A team led by Nataliya Kosmyna at the MIT Media Lab, with Pattie Maes among the authors, posted "Your Brain on ChatGPT" to arXiv on June 10, 2025 (revised December 31, 2025). Fifty-four participants wrote essays over three sessions in one of three groups: with a language model, with a search engine, or with no tools. Eighteen took part in a fourth session in which the tool groups were switched.
The authors report that EEG connectivity was strongest in the no-tools group and weakest in the language model group, that language model users felt the least ownership of their essays and had trouble quoting them, and that those who moved from the model to no tool showed reduced connectivity. The arXiv page does not state that the paper has been peer reviewed. The sample was small, and the task was essay writing, so the findings fit best as a reason for further trials, not as a verdict for classrooms.
Reading these findings carefully
The two sources differ in kind. The Pew data say how common a behavior is. The MIT preprint hints at a mechanism, that handing off the drafting step may reduce the effort that builds memory. Neither measures whether a student who uses a chatbot ends up knowing less on a later test. Experiments that do measure that exist for specific subjects, and results depend on how the tool is designed, so a single headline should not decide school policy.
Checks a school can build around the downsides
- Ask students to show drafts and notes, and grade some work done without the tool.
- Name which tasks allow AI and which do not, task by task.
- Survey local students on actual use before writing rules, since national numbers may not match.
- Treat peer-review status as a question to ask of any study cited in a policy memo.
This article reports survey and research findings. It is general information, not legal or professional advice, and it does not recommend a tool or product.
FAQ
How many students use AI for schoolwork?
In Pew's fall 2025 survey of U.S. teens, 54 percent used chatbots for schoolwork. Canadian official figures were not found, and commercial surveys differ by method, so they are not cited here.
Does the MIT study prove that ChatGPT makes people less intelligent?
No. It was a preprint with 54 participants and an essay task, and it reports brain connectivity and recall measures, not intelligence or later exam results.
Do students think AI helps?
Many do. About a quarter of U.S. teens in the Pew survey called chatbots extremely or very helpful for schoolwork, and another quarter called them somewhat helpful.
Is student cheating with AI measured?
The Pew figures are what teens perceive about their peers, not counts of confirmed cases.
Related reading
- Limits of AI in education
- Possibilities for AI in schools
- Ethical questions raised by AI
- How AI is used in learning
More in AI in education.
Last verified: October 2026
Sources
- How Teens Use and View AI, Pew Research Center, February 24, 2026
- Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, Kosmyna N et al., arXiv:2506.08872, 2025