A Minneapolis teacher banned phones and laptops for a school year. What her students reported is a working precedent.
I teach in Shanghai, which means most of my classroom decisions about AI are local. The school has a policy. I have a syllabus. The decisions about what students outsource sit between the two, in the assignments I give.
Last week, that space got a new piece of evidence.
Maureen Mulvaney teaches AP Literature at Washburn High School in Minneapolis. At the start of the 2025–26 school year, she banned phones and laptops from her classroom. Students were required to do all of their work with pencil and paper, swapping digital PDFs for physical books. Vice’s Luis Prada covered her experiment last week.
The results, by Mulvaney’s count, were striking. In September, 46% of her students said they felt confident in their reading. By February, that number had climbed to 95%. Students who started the year struggling to produce a single paragraph by hand were turning in five or six pages. Nearly 80% reported that thinking and organizing ideas felt easier on paper than on a screen. Some students said their screen habits at home had dropped too, almost as a side effect.
A few honest caveats up front. This is one classroom, the numbers are self-reported, there is no control group, and Mulvaney is the experimenter and the teacher and the reporter all at once. As causal evidence about AI in schools, this is not very strong.
As a working precedent, it’s worth a closer look.
What Mulvaney’s experiment lines up against
A 30-month study of 26,811 Chinese secondary students, published earlier this year by Strömberg, Lei, and Wu, found that students using generative AI raised their homework scores by 18% and finished 30% faster. Their monthly exam scores fell 20% within six months. On entrance exams that decide college placement, the gap grew to 18–24%, and the full hit only surfaced after about two years. Eighty percent of users showed the same pattern: fast completion, strong homework marks, and weaker performance once the tool was gone.
That study is the closest thing we currently have to causal evidence on the AI-in-school question. The damage concentrates in students who were using AI as a shortcut, even when nothing about that use looked like cheating.
Mulvaney’s classroom is the inverse experiment: remove the tool entirely and watch what comes back. What came back, in her telling, was reading confidence, writing stamina, and a noticeable shift in how students described their own attention.
The “just hoping” gap
Vermont issued AI guidance for schools in January 2026. The document is 50 pages, built around a developmental framework, with named stages of progression mapped to grade bands. It’s the most carefully sequenced state document I’ve seen so far. It’s also voluntary.
North Carolina issued guidance in 2024 that Vermont built on, but Vermont’s version is the rare one. Most other US states haven’t published anything beyond general guidance about AI in schools.
The rest of us are just hoping it turns out fine.
Mulvaney didn’t wait for a policy. She made one assignment-level decision and then made it again every day for a school year. Whatever the eventual research shows about pen-and-paper-only AP classrooms, what she has is a year of measured classroom outcomes against a tide of “just hope” everywhere else.
What this means for teachers
You don’t have to do what Mulvaney did to take something useful from this. The total ban is one end of a gradient. The other end is unstructured device use during every assignment. Most teaching happens between those two.
What her experiment forces is whether students are required to do the cognitive work AI tends to absorb before the tool becomes available: the slow part of reading, the muscular part of writing, the figuring-out part of figuring out. If a student writes their first draft on paper and then types it up and uses AI for polish, the cognitive work has happened. If they generate the first draft in ChatGPT, it hasn’t.
The Strömberg study is asking, in measurement form, whether the cognitive work happened. Mulvaney is asking, in classroom form, what it costs to be sure it did.
Neither has a complete answer yet. Together, they have the shape of one.
Takeaway for Teachers
This week, pick one assignment where AI use is currently optional and turn it into an assignment where the first 20 minutes are paper-only. No phones, no laptops. Let students sketch the structure of what they’re about to write by hand. Then, if you want, let the tools come back in for the typing, the cleanup, the research.
You will see in two weeks whether their drafts have changed.
Mulvaney did the big version. The small version is the one most of us can run on Monday.
If you’re thinking through questions like this one — what to require of students before the tool comes in, and how to know whether the cognitive work has happened — my book goes deeper. The AI Doesn’t Know Your Students is available on Amazon and at shouldiuse.ai/book.
David Jacobson is a high school history teacher in Shanghai. He writes about AI and education at shouldiuse.ai and is the author of The AI Doesn’t Know Your Students.
