Middle Schoolers Got Free AI Tutors. Most Used Them to Ask for Jokes, Not Help.
A two-year study found kids barely used an AI tutor for real learning, while curriculum-built AI reading tools show promise — the difference seems to be how tightly a teacher controls the tool.
What happened: A two-year study across 18 Tennessee middle schools gave students free access to Khan Academy's Khanmigo AI tutor during the school day. Researchers found access was "nearly universal but engagement was thin": most kids ignored it for actual coursework, instead sending off-topic prompts like requests for jokes or an essay on pizza history, or trying to trick it into handing over answers. Math scores still rose modestly among users. Separately, some districts in Texas and Tennessee report stronger reading-test growth using AI tools built directly into literacy curriculum rather than handed to students as open chatbots — a very different setup from the one in the engagement study.
Why it matters: Middle school is where many kids hit a "comprehension wall": they can sound out words but lose meaning once texts get harder or vocabulary and background knowledge run thin. Only 30% of U.S. eighth graders read proficiently, the lowest mark in 30 years. One teacher can't give two dozen-plus students individualized, real-time feedback, so AI is being pitched as the way to deliver that at scale. But if students won't engage with the tool unless someone makes them, the promise of automatic, scalable support quietly depends on exactly the human attention it was supposed to free up.
Two very different tools: The Tennessee chatbot was open-ended: students could type anything, and many used that freedom to avoid work rather than do it. "The same technology that can be used as a good tutor, can also be used to make your life easier," said researcher Philip Oreopoulos — and easier usually means less thinking. Curriculum-embedded literacy tools work differently: they diagnose where comprehension breaks down in assigned texts, ask a targeted follow-up question, redirect students to reread a specific sentence, or require a revised answer before moving on. Stanford researchers separately found AI tutoring works best when built for teachers to use with students, not handed to students alone.
The rollout: Khan Academy CEO Sal Khan called the findings a reminder of how much motivation matters, not a surprise to most educators. Advocates for AI literacy tools argue districts should ask hard questions before buying in: does the tool keep kids on grade-level text, require them to revise their thinking, and give teachers visibility into where struggles happen — or does it just isolate students on a screen? The emerging consensus from both camps is the same: AI tutoring's success hinges less on the software and more on whether an adult is actively steering how it gets used.