Technology in schools
Coding, design, and how apps are built to hold attention belong in every school. A Canva template for the science fair or a chatbot for the term paper often does the part of the work that was the lesson.
Try this before you read on. Open any AI chatbot and type what a 5th grader would: help me with my science fair project about plants. Then count the decisions it leaves for the kid. The question, the steps, the materials list and the poster layout usually arrive in one reply, with an offer to do more.
For an adult at work, that reply is the product. For a 10-year-old, those decisions were the project.
Two different things hide under the phrase “tech in schools.” Teaching tech means teaching kids how technology works: how a computer follows instructions, how an app gets designed, how a feed decides what to show you. Teaching with tech means using a tool to do schoolwork that could be done another way. The first is a subject, and most schools teach too little of it. The second is a method, and it needs a reason every time.
A 4th grader who builds a science-fair board with scissors and glue decides what goes where, runs out of room, and starts over. A 4th grader in Canva picks a template. Both boards look fine on the day, but only one kid did the layout. Research works the same way. A 9th grader who asks a chatbot for sources gets a tidy list and a summary, and never finds out that half the sources disagree. Working out why they disagree was the assignment. Teachers call that part productive struggle, and most tools are built to remove it.
Dan, who started Screenwise, puts his own family's rule on our story page: “No YouTube, no Minecraft, no Roblox, no TikTok, no ChatGPT. That works for us (for now!) — but that model doesn't have to work for you.” Your family can land anywhere on that. The school question is narrower: which work should a kid do by hand, and which technology should they learn to understand?
The rule this page argues for is short. Teach tech on purpose, and teach with tech only when you can say what the tool adds that the struggle doesn't.
Teaching tech
Teaching with tech
Build a game in Scratch and fix the bug that lets the cat walk through walls
Play a math game that pays out coins for right answers
Redesign a real app screen and test it on three classmates
Make the science-fair poster from a Canva template
Work out why YouTube plays the next video on its own
Watch a YouTube video about the water cycle
Learn how a chatbot is trained, and catch it inventing a source
Ask a chatbot to find sources for the term paper
Learn to type, name files, and search with the right words
Do the same worksheet on a Chromebook instead of paper
The right column is not banned. Each item needs a reason the left column doesn't.
How sure we are
In a randomized trial with nearly 1,000 high school students, a plain ChatGPT-style assistant raised practice scores 48%, then students scored 17% lower on the exam once it was gone, and they didn't notice they had learned less. A tutor version that gave hints instead of answers avoided the harm. Separately, reading comprehension is better on paper than on screens across more than 170,000 readers, and the gap has grown over time.
The studies, sorted by what they foundNationally, 54% of US teens have used a chatbot for schoolwork and 1 in 10 use one for all or most of it (Pew, 2026). We ask Screenwise parents: "ChatGPT / Claude / Gemini: does your kid use any of these?" Of 580 families, 16% say yes for kids in pre-K to 2nd grade, 34% in 3rd to 5th, 58% in 6th to 8th, and 63% in high school. ChatGPT's own terms set 13 as the minimum age and Claude's set 18, so many of the 8-to-11-year-olds in that count are using tools set for teenagers or adults. Google is the exception: it offers Gemini to under-13s through parent-supervised Family Link accounts.
Teaching tech versus teaching with tech, the six questions, and the grade-by-grade defaults below follow from the research. No study has tested them as a set. Treat them as a starting position for your school, and change them when your own check-ins say to.
An old argument
In 1966 and 1967, years before most adults had touched a computer, Seymour Papert and three colleagues designed Logo, a programming language for children. Soon after, at MIT, kids were using it to steer a small robot “turtle” across the floor, drawing as it went. When the drawing came out wrong, the kid had to work out which instruction was wrong. In his 1980 book Mindstorms, he described what schools were doing with computers instead:
“In many schools today, the phrase ‘computer-aided instruction’ means making the computer teach the child. One might say the computer is being used to program the child. In my vision, the child programs the computer.”
People who worked with him built on that. Mitch Resnick's group at the MIT Media Lab released Scratch in 2007. More than 150 million young people have used it since. It is still the best first step into computer science most kids will get.
Papert's test sorts almost every tool on this page. In Scratch, the kid tells the computer what to do and fixes it when it goes wrong. In a math app that pays out coins, or a chatbot that drafts the report, the computer does the thinking and the kid follows along. The chatbots are far more capable than anything Papert saw, and his question still applies to them.
Teaching tech
Five subjects, starting in kindergarten. Much of it happens on paper, and the rest is screen time spent studying the screen.
Every app a kid uses was designed by people making choices: what the button says, what shows up first, what is hard to find. A kid who has designed a screen themselves starts to see those choices everywhere.
How a computer follows instructions, how to break a big problem into small steps, and how to find the bug when it does the wrong thing. It starts on paper in kindergarten and ends in real code.
The plain basics: typing, saving and naming files, folders, keyboard shortcuts, searching with the right words, knowing where a download went. Kids raised on tablets often reach middle school without them.
Find a real problem, talk to the people who have it, build a rough version, watch it fail, and improve it. Most of this happens away from a screen, which is why kids learn so much from it.
Streaks, autoplay, infinite scroll, loot boxes, the red notification dot. Each one is a design decision made by someone paid to increase time spent. A kid who can name the trick has a defense against it that no filter gives them.
Teaching with tech
Most of the technology in a school day is this second kind: Chromebooks for worksheets, slides for presentations, adaptive apps for math practice, and now chatbots for research and writing. Each arrived with good intentions, and each tends to take a hard part of the work and make it easy.
Sometimes that is the right call. A simulation of planets in orbit shows something paper can't. A text-to-speech tool opens a book to a kid with dyslexia. A video call brings a scientist from another continent into the room. In each case the tool adds something the kid could not get any other way.
The trouble starts when the hard part being removed is the lesson. Laying out a poster teaches planning. Finding sources teaches judgment. Writing a first draft teaches thinking. A tool that does those for a kid produces a better-looking result and a kid who learned less. The classroom research points the same way: for most of the school day, teachers and paper outperform screens.
The struggle has to be within reach, though. A review of 53 studies found that letting students attempt a problem before being taught helped them understand it better, but for 2nd to 5th graders the order flipped and teaching first worked better. Young kids learn from hard work they can finish, with an adult close by. Frustration they can't get through teaches them nothing.
What is the kid supposed to get better at, and does the tool do that part for them?
Ask it of every assignment that opens a laptop.
AI in schools
The chatbots adults use every day are tuned to be excellent employees. Their instructions boil down to this: work out what the person wants, do as much as you can on your own, check in only when you're stuck, stay encouraging, and get the job done.
For a knowledge worker, that is the product. An adult with an AI assistant can research, write, build, and automate things that used to take a team. Plenty of parents reading this use one all day.
Give the same assistant to a 12-year-old and every one of those instructions works against them. Working out what they want means they never have to say it clearly. Doing as much as possible means doing the thinking. Checking in only when stuck means the kid never gets stuck. Staying encouraging means the weak idea gets praised. Getting the job done leaves nothing for the kid to do.
The agreeable part is not an accident. Researchers at Anthropic, which makes Claude, found in 2023 that the leading assistants tend to tell people what they want to hear, because the ratings used to train them reward it. In April 2025 OpenAI rolled back a ChatGPT update after users found it too flattering.
A kid doesn't need an employee. They need a coach, and a good coach sometimes says no.
The studies so far
Answers on tap raise homework scores and lower test scores.
Nearly 1,000 high school math students in Turkey: +48% on practice with a plain chatbot, then 17% lower on the exam without it. Students mostly asked it for the answer, and did not notice they had learned less. (Bastani et al., PNAS, 2025)
A better essay is not a better writer.
University students revising with ChatGPT improved their essays more than students helped by a human expert, but learned no more and transferred no more to new tasks. (Fan et al., British Journal of Educational Technology, 2025)
Tutors built to withhold answers can work.
A Harvard physics tutor designed to walk students through each step roughly doubled learning gains over an active-learning class, in less time. A teacher-run after-school program in Nigeria using a chatbot raised English scores 0.23 standard deviations in six weeks. (Kestin et al., Scientific Reports, 2025; World Bank, 2025)
Only if kids use them.
Two years, 18 Tennessee middle schools: a coach-style tutor helped about as much as ordinary online practice. The median student messaged it in 17% of the sessions where they made a mistake. (Oreopoulos and Low, NBER, 2026)
The pattern shows up worldwide.
In the OECD's 2025 PISA survey, students who used AI to draft, summarize or research scored about 20 points lower in science, around a year of school. Among frequent users, those taught to judge AI output did better. This is a correlation from a survey, not an experiment. (OECD, 2026)
Built for adults at work
What a kid's AI should do
Works out what you want and fills in the gaps
Asks what the kid thinks first, and waits for the answer
Does as much as it can on its own
Does the smallest useful piece, then gives the work back
Checks in only when it is stuck
Checks in at every step and asks the kid to take the next one
Encouraging and agreeable
Plain about weak spots: "your second paragraph contradicts your first"
Finishes the job
Stops before the part the kid is there to learn
Refusing to write the essay is the easy case, and most tutor modes already do it. The better version won't move on until the kid has done the part of the work that happens out in the world.
A 7th grader wants help building an app that reminds kids to feed their pets. A good AI coach does not write code yet. It asks who has this problem and sends the kid to talk to three friends with pets. The code conversation starts after the kid reports back what they heard.
A 5th grader asks for a report on the creek behind the school. The coach helps plan what to measure and who to call at the county parks office, then waits until the kid has been to the creek with a notebook.
A 10th grader wants to write a persuasive essay on later school start times. The coach asks them to survey ten classmates and interview one teacher before it helps with an outline.
Adults can skip the legwork because they have done it a thousand times, and a kid hasn't done it yet.
Several products now ship a tutor or study mode, including Khanmigo, ChatGPT's study mode, and Claude's learning mode. They are a real step forward. Many still sit one tap away from the default assistant, so the setting has to be locked by how the school configures the tool, not left to a kid's willpower at 9pm.
Here are instructions a school can paste into any AI tool it sets up for students, or send to a vendor as the standard to meet. Fill in the grade and the skill.
Instructions for a kid-facing AI coach
You are a learning coach for a student in grade [X]. Your job is to help them get better at [the skill this assignment builds]. It is not your job to finish their work. - Before you help, ask what they think and what they have tried. Wait for their answer. - Give the smallest hint that gets them moving. Do not write sentences, code, or answers they could produce themselves. - After each step, give the next step back to them. - Be direct. If an idea is weak or an argument has a gap, say so and ask how they would fix it. - If the work needs real-world steps (an interview, a measurement, reading a source, testing an idea on people), ask them to do it and report back before you go further. - When you state a fact, say where it comes from and how they could check it. - If they ask you to do the work for them, say no, say why in one sentence, and offer the next small step.
Decision framework
For a teacher planning one assignment, a principal renewing a license, or a parent looking at tonight's homework.
If the tool is the subject, the screen time is the lesson. If the tool is the method, keep going down this list.
Name the skill in one sentence. If nobody can, the assignment is not ready, with or without a tool.
Tools are built to remove effort. The question is whether the effort they remove is the lesson.
Paper first, tool second keeps the thinking in the kid. The tool can polish what they already made.
Ed-tech is sold on engagement, and engagement is a vendor's number. Learning is the school's.
Many tools set their own minimum age, and every tool collects something. The school should be able to say what.
For question six, the parent's rights and the two requests worth making in writing are on Screens in Schools.
Defaults by grade
Fair objections
They will need it.
The skill that matters at work is judging and directing what an AI produces, and you can only judge an essay you could have written. That is teaching tech, and the grade table above starts it in middle school. What waits is handing a 10-year-old the tool that writes the essay.
Partly right.
A 1986 review of 79 studies found calculators used alongside regular math teaching helped students at almost every grade. The exception was 4th grade, the year arithmetic is being built: sustained calculator use there held basic skills back. The same rule works for Canva. When the science is the lesson, a quick deck is fine. When planning the layout is the lesson, sketch it on paper first.
One-on-one tutoring is one of the strongest results in education research.
Benjamin Bloom's 1984 study found tutored students scored about two standard deviations above classroom peers. A 2020 review of 96 randomized tutoring studies found a smaller effect, 0.37 standard deviations, still large for education. Those tutors asked questions and made the student do the work. A chatbot that gives the answer when asked is a different thing. The AI tutors that work are built like those tutors: a Harvard physics tutor that walked students through each step, instead of solving problems for them, beat a strong in-class lesson. Even then, kids have to use it. In a two-year trial in 18 Tennessee middle schools, a coach-style tutor added about as much as ordinary practice, because most students rarely asked it anything.
An app labeled educational can be a worksheet with a progress bar, and many assign hours a week. The label settles nothing. The design does.
The apps schools assign, scoredAn app that pays out stars for dragging blocks into the right slot is practice, not computer science. Computer science starts when a kid designs something of their own and fixes their own bug.
A school that removes every screen also removes Scratch, typing, and the unit on how feeds work. Kids then learn technology from TikTok and the group chat, with no teacher in the room.
A tool the school approved last year adds an AI feature in October, and nobody re-checks it. Any new AI feature, store, or terms change sends the tool back through the six questions.
Implementation
Weeks 1-2
Every app, site, device, and AI feature. Sort each into three piles: teaching tech, teaching with tech, and admin (attendance, grades, messages home). Most schools have never seen this list in one place.
Weeks 3-4
Left: what we teach about technology at each grade. Right: what we use technology for, with the reason next to each tool. Send it to families.
Weeks 5-6
Most of it is free: CS Unplugged activities, ScratchJr, Scratch, and the Code.org courses. Add one unit a year, from 3rd grade up, on how apps are designed to hold attention.
Weeks 5-8
Teachers can use AI for their own work. Students learn about AI as a subject, by grade band. Any AI tool a student uses runs in a coach mode with written instructions, and follows the tool's own age rules.
Week 8
Give parents a ten-minute assignment and have them do it the way their kids do: with the tool, then without it. That hour answers more questions than a policy PDF.
Weeks 9-18
Run the six-week check below at least twice before the plan spreads to the next grade.
Ask for the tool list and what each tool is for. Most schools will send it, and the two questions in the email do more than a debate about screen time.
To your principal
Subject
Which tools does [school] use this year, and what is each one for?
Body
Hi [Principal's name], I'm the parent of a [grade] student in [teacher]'s class. Could you share the list of apps, websites, and AI tools students use this year? For each one, I'd like to know two things: 1. Is it teaching kids about technology (coding, design, how apps work), or is it a tool for doing other schoolwork? 2. If it's the second kind, what skill is the assignment building, and does the tool do that part for them? I'm happy to help turn this into a one-page summary for other families. Thanks, [Your name]
Check-ins
One test matters more than the rest: can the kid do it without the tool?
Teachers
The school
District, PTA, or community
Whoever owns the list
Most science fairs happen in the spring. The board a kid builds by hand will be a little crooked, and the title won't fit on the first try. They will glue it down anyway, and they will be able to tell you why every piece is where it is.
Tonight, ask your kid what they made this week that no computer helped with, and what they learned about how a computer works. If both answers are “nothing,” the email to the principal above is a good place to start.
For your family
Sign in to personalize this guide with data from families in your school, city, and community
See your own school's numbers.
The five-minute Screens in Schools survey collects which devices and AI tools your kid's school uses and how many hours they take, and shows it next to other families at the same school.
Take the Screens in Schools SurveyGo deeper
The AI tools in your kid's classroom
Which ones tutor and which ones give away the answers.
When AI writes the essay
What to do when you find out, and what to set up so it stops.
ChatGPT's study mode
What it changes, and how easy it is to switch off.
MagicSchool AI, for teachers or students?
Teacher tools that end up in front of kids.
Teaching kids to question AI
Catching made-up facts and flattery.
Screens in schools: the research
Reading, attention, device programs, and your rights as a parent.
Ed-tech, scored
The apps schools assign, each with a WISE score.
Parental controls on school laptops
What you can and cannot control on a school device.
School device monitoring
What the school sees, and what you can ask about.
Family figures come from Screenwise survey answers from 580 families, collected October 2025 to September 2026. Everything else links to the study or official source.