The family AI setup for back to school
Seven in ten teens already use AI and most parents find out later. One calm evening of setup covers the whole school year.
The most common first week of school in America is the one in mid-August. In Pew Research Center’s analysis of district calendars, roughly 70% of the country’s 46.7 million public school students were back in class by late August, with the South starting as early as the first week of the month and New England holding out past Labor Day. Wherever your district falls, the syllabus is about to land on the kitchen table.
This year, one more thing belongs on the back-to-school list, somewhere between the backpack and the dentist form: deciding how your family uses AI. Not because it’s trendy, but because the kids have already decided without you. Common Sense Media’s survey of 1,045 parent-and-teen pairs found seven in ten teens have used generative AI, while only 37% of their parents knew. The good news: getting ahead of it is one calm evening of setup, it costs nothing, and the deep thinking is already done in three posts we published this year for students, parents, and teachers. This is the checklist that stitches them together.
Do it before the syllabus lands
Families mostly meet AI mid-crisis. It’s 9:40pm, the essay is due tomorrow, and a parent is discovering in real time that the paragraph on the screen wasn’t written by their child. Rules improvised in that moment come out as bans, and bans lose to the numbers above. That same Common Sense survey found 40% of teens had used AI for school assignments, and 46% of those did it without the teacher’s permission, mostly because nobody had said a word about what was allowed.
Don’t wait for the school to lead. In the same survey, 83% of parents said their school had not communicated with families about generative AI at all. The syllabus may eventually carry an AI policy line, and where it does, the syllabus wins. But the family conversation, the accounts, and the habits are yours to set, and the cheapest time to set them is the quiet week before assignments start, when nobody is defending a deadline.
The student setup is one rule, not a lecture
The temptation is a family seminar on academic integrity. Skip it. The whole thing compresses into one rule: AI as a tutor is always fine, AI as a ghostwriter never is. A tutor asks you questions, explains what you got wrong, and leaves you a little sharper. A ghostwriter hands you a finished thing to submit and leaves you exactly where you were. Every gray case a school year can produce sorts itself with one question, asked before hitting enter:
Worked examples make the rule concrete. “Quiz me on chapter 4” passes. “Write the introduction to my essay” fails. “Give me feedback on my draft introduction” passes, because the rewriting stays in the student’s hands. This isn’t about scaring anyone straight; used as a tutor, AI is genuinely excellent, and pretending otherwise burns credibility. With 86% of students worldwide reporting they use AI in their studies in the Digital Education Council’s 3,839-student survey, the realistic goal is a student who uses it well, not one who doesn’t use it.
The rule isn’t just etiquette; it tracks how learning actually responds to AI. In a PNAS field experiment with nearly 1,000 high schoolers, students given a plain chatbot during math practice scored 48% better on the practice, then 17% worse on the exam, where the AI was gone. Students given a tutor-mode version, which offered hints instead of answers, improved 127% on practice with no exam harm at all. Same model, same students, same material. The only difference was whether the AI was allowed to hand over the deliverable. That’s the ghostwriter cost, measured.
Two habits complete the setup. First, read the AI section of each syllabus when it arrives; teachers’ rules differ, and the syllabus always wins. Second, when in doubt, disclose. A one-liner like “I used AI to quiz me on the readings; the writing is mine” has never made a situation worse. The full students post has the gray-zone table and the five tutor-mode prompts if the one-evening version sparks questions.
The parent setup pays back more hours
Here’s the part the back-to-school articles skip: the parent gets more out of one evening of AI setup than the student does. A Journal of Marriage & Family study of 3,000 US parents found mothers carry 71% of the household’s mental-load tasks, the invisible planning and scheduling that a school year multiplies. Much of that load is text: permission slips, teacher emails, camp letters, the 900-word PTA newsletter with one date buried in it. Text is the thing AI is best at.
The setup is a chat app on the parent’s phone and three reflexes. Photograph any form and ask what needs signing, paying, and packing, and by when. Draft the awkward teacher email by giving the facts and the tone you want. And on Sunday evening, paste the week’s mess in and ask for the conflicts and the night-before prep. The parents post carries the full prompt kit, from birthday-party planning to the weekly reset. It’s worth adding that the person on the other side of the desk is running the same play: teachers who use AI weekly report saving 5.9 hours a week, per Gallup’s survey of 2,232 US teachers, and our teachers post maps where those hours come from.
One rule keeps the parent setup safe: the kid’s name stays out of the chat. Swap real names for “my 8-year-old,” leave out the school’s name, and keep health details offline. Consumer chat apps keep what you type under retention policies most people never read; we read them so you don’t have to. The redaction habit costs nothing and covers most of the risk.
Give the family computer a private helper
The most underrated piece of the family setup is the computer already in the living room. A modern laptop with 8 to 16GB of memory can run a small open AI model locally, free, with no account, and with one property no cloud chatbot offers: homework questions never leave the house. A ten-year-old’s reading struggles, a teenager’s essay drafts, the questions a kid is embarrassed to ask out loud; none of it belongs in a stranger’s retention policy, and with a local model none of it goes there.
Honesty requires saying the local model will be less capable than the big cloud ones. For explaining fractions, quizzing chapter 4, and hinting at the next algebra step, the gap rarely matters. Picking a model sounds technical and isn’t: our flowchart picks one in five questions, and the RAM guide tells you what your machine can actually hold. In practice the install is one app and one download, about the size of a video game update, and the thirty minutes budgeted for it in the checklist includes the test drive. We build a desktop app that runs models this way, so we’re biased and telling you so; the pattern works in any local-first tool.
There’s a quieter reason to make the private helper the family default. A student who knows the chat stays on the machine asks the embarrassing questions, the “I still don’t get why negative times negative is positive” questions, that they would never type into an app that keeps a transcript under their account. The best tutor is the one the kid actually talks to.
Syllabus week runs on five prompts
The first week of school produces a specific pile: syllabi, supply lists, schedule grids, club signups. Five prompts, agreed on during the setup evening and pinned somewhere visible, turn the pile into a system. Three belong to the student and encode the tutor rule; two belong to the parent and eat the paperwork.
None of these needs magic words. A good prompt is a clear brief: what you’re handing over, what you want back, and what the AI should not do. The “don’t solve it for me” clause is the load-bearing part of the student prompts; it converts the same tool from answer machine to tutor. Write the five onto a card, tape it next to the family computer, and the rules survive the semester better than any lecture.
Buy nothing in week one
Back-to-school season is engineered for spending, and AI companies know it. Resist. Every piece of this setup runs on free tiers: the chat app on the parent’s phone, the student’s quiz-me sessions, the local model on the family computer. Free-tier limits exist, but a family discovering AI will not hit them in September, and if a limit ever pinches, that’s the moment to decide, with a month of real usage as evidence.
The test we suggested in our subscriptions post applies doubly to school tools: if canceling it a month later would leave you relieved rather than panicked, it should never have been a subscription. A $20-a-month plan bought in the week-one enthusiasm becomes a $240 line item that outlives the enthusiasm by two semesters. Start free, upgrade the one thing the family actually misses, and skip the AI-branded study apps entirely until a teacher names one.
One evening, whole year
The whole setup, on one page: agree on the tutor-not-ghostwriter rule and its one-line test. Put a chat app on the parent’s phone and practice the form-photo reflex. Install a local model on the family computer for the questions that should stay home. Write the five prompts on a card. Spend nothing. It’s about an hour of actual work, and the payoff arrives in November, when the essay crisis that derails other households is, in yours, a kid asking the draft critic for feedback at a reasonable hour.
Rules set calmly in August beat rules improvised at 9:40pm in October. That’s the entire trick, and this is the week to use it.
Disclaimer: This is general study guidance, not academic-policy advice. Your institution’s academic-integrity policy and each instructor’s rules are the documents that control; where this post and your syllabus disagree, the syllabus wins. When in doubt, ask before you submit.
