AI in Schools: 70% of Teens Use It for Schoolwork. 70% Say No Teacher Has Discussed It.

44% of teens who use AI for schoolwork have run into a school network that blocks it, and 59% of them finished the assignment on a personal device. Those figures come from a survey of 1,017 American teenagers between the ages of 13 and 17, published on August 18, 2026 by the Youth AI Safety Institute at Common Sense Media. They arrive attached to two larger numbers that describe the same classroom. About 70% of teens now use AI for schoolwork, and 70% say no teacher has ever discussed safe or responsible use with them.

63% of teens use AI specifically to get answers to schoolwork. What happens to the answer after that is where the education either occurs or evaporates.

Another 41% use writing assistance like Grammarly, a category most adults stopped thinking of as artificial intelligence years ago. And 66% report that the model helped them understand material they had been struggling with, which is the reason a prohibition was never going to hold in the first place.

The number worth sitting with is 39%

39% of these students say they feel they are missing out on learning when they use AI on an assignment. Nobody prompted that admission with a lecture about integrity. They were handed a survey, they graded themselves, and 4 in 10 reported a loss they could feel while the score on the paper stayed perfectly respectable. Another 38% said they generate fewer of their own ideas than they used to. Teenagers are describing skill atrophy in their own vocabulary, years before any professor, admissions officer, or employer will describe it for them.

Ashley Green, the education technology manager at Napa Valley Unified School District in California, has been asking students what keeps them away from these tools. Their reason had nothing to do with ethics.

"These kids are saying, 'Oh, we're never going to use it because we're so scared that we're going to get in trouble.'"

Ashley Green, education technology manager, Napa Valley Unified School District, describing what students tell her

That fear is a curriculum outcome nobody designed. A student who is frightened of the tool avoids admitting she used it, and a student who cannot admit she used it never gets taught to check what it told her. Blocking the tool changes where the homework happens. Teaching the check changes what the homework is worth.

Every school system is making a four-year bet

A district that sets a graduation requirement today is wagering that the skills behind it will still matter when this year's freshmen walk across the stage. That wager was reasonable for most of a century, because the working world moved slowly enough for a curriculum to stay current with it. The curriculum half-life is now shorter than the credential it belongs to, and the students in this survey are already living inside the difference.

They are also the freshman class of 2030. They will arrive on campus having done homework alongside a model since middle school, and the universities admitting them are placing a four-year bet of their own on what those students can actually do unassisted. Purdue has moved on its version of the question by writing AI into its graduation requirements. A 3-year study at Babson College found that the same students' deepest worry migrated from being accused of cheating to losing their own voice. Both institutions are reading the evidence that this survey has now placed in front of every district in the country. Every story I follow in AI in education returns to the same arithmetic: the institution commits for four years, the tools revise themselves in four months, and the students reconcile the difference on their own time.

The cheapest thing to fix is the conversation

John King, former U.S. Secretary of Education, framed the question directly: "What are we doing to equip students with the skills to navigate AI and use it effectively?"

That question has a less expensive answer than most people assume. Teachers received this technology mid-year, with no curriculum attached to it and no budget behind it, and many are already doing the work without anything official to point to. A written classroom rule about when AI is welcome costs a district nothing to distribute. One class period spent checking a model's answer against a primary source teaches the single skill every one of those 63% will need for the rest of their working lives.

I advise the California State University system's AI Working Group, and the question that surfaces most often in those conversations has almost nothing to do with detection software. It concerns what a passing grade should certify by the time these students arrive.

If you are a parent, your move this week is smaller and better. Ask your kid to show you the last assignment where a model helped, then ask how they knew the answer was right. The second question is the entire education.

Yvette Renteria, chief program officer at Common Sense Media, described what she wants educators to keep hold of: "Educators embrace gray areas. That's the beauty of learning."

The gray area is precisely where the teaching lives, and it survives only in classrooms where a student can say what she used without bracing for consequences. The student who moved to her own phone finished her homework either way. Whether anyone taught her to check it is the part still in play, and it is the part adults control. Take that question into your next conversation about AI policy and watch who reaches for the answer first. I am easy to find.

Here is what makes Alex a credible voice on this topic: Alex Goryachev advises the California State University system's AI Working Group on AI governance and is Innovator-in-Residence at Tulane University, which places him inside the same question this survey puts to K-12: what a credential should certify once every student has a model in their pocket.

Bringing this question to your district, campus, or leadership team? Book a conversation →