Print this, forward it, or hand it to whoever owns the policy. Both the brief below and the policy below it may be copied, edited and reprinted without asking and without credit.
What is actually happening
Most students are already using AI for schoolwork, and most teachers have been given no formal position on it. In a nationally representative sample of 2,069 US public school teachers, only 18% had received written or official guidance on classroom AI use, and 34% had received none at all (RAND / Gallup / Walton, fielded 9 Feb to 2 Mar 2026). Separately, 86% of children aged 9 to 17 now use AI, and 85% of those use it for schoolwork or homework (Common Sense Media, 8 Jun 2026). The gap is not whether students use it. It is whether anyone told them, or their teachers, what is allowed.
What the evidence does and does not support
AI tutoring has real, modest, measured effects: a six-week randomised trial in Nigeria found a 0.23 SD gain in English from AI-assisted tutoring (World Bank, 2026). It does not support strong claims either way about teaching AI literacy itself: no study has tracked whether that instruction changes student judgement a year later, and the most-cited harm study tested 54 adults, not children, and was never peer reviewed (MIT Media Lab, preprint, 2026). Treat both the promise and the panic as unproven.
Three questions your department should be able to answer
- Which assignments allow AI, and which explicitly do not?
- What does a student have to disclose, and who checks it?
- Where is the guidance written down, and who has actually read it? Most teachers have not seen one (RAND / Gallup / Walton, 2026).
What it costs to get there. One meeting to draft a position, one to adopt it. Nothing below requires a vendor, a budget line or a new subscription.
A draft policy to argue with, not adopt
This is a starting point, not a template. Take it into a staff meeting, disagree with parts of it, and rewrite whatever does not fit your school before anyone signs off on it.
Version, 5 August 2026
Draft classroom AI use policy
Which assignments allow AI
Each assignment is marked AI-permitted or AI-free by the teacher who sets it, in writing, before the assignment is given. Where no mark is given, the default is AI-free. AI-permitted work still requires the student's own judgement, structure and final wording.
What disclosure looks like
Any student who uses AI on AI-permitted work states, in a line at the end of the submission, what tool was used and for which part of the process, for example brainstorming, a first draft, or feedback on a draft. Disclosure is not required for spelling and grammar tools already in classroom use.
What happens when a student goes outside these lines
The first response is a conversation with the student, not an automatic grade penalty. A teacher's own judgement of the student and the work counts for more than an AI-detection tool, which produces false positives and flags some groups of students more than others. Repeated or deliberate breaches follow the school's existing academic integrity process.
The one hard rule
Never enter a student's name, ID number, grades, health information, or any other personal information into an AI tool that is not approved and contracted by the school. This rule has no exceptions and no judgement calls.
Sources
Full evidence write-up and methodology:the research page.
- RAND / Gallup / Walton, "Teaching for Tomorrow: Closing the Expectations Gap," fielded 9 Feb to 2 Mar 2026, 2,069 US teachers.news.gallup.com
- Common Sense Media, inaugural census of AI use by tweens and teens, released 8 Jun 2026.commonsensemedia.org
- World Bank, "From Chalkboards to Chatbots" (Nigeria RCT), 2026.documents.worldbank.org
- MIT Media Lab, "Your Brain on ChatGPT," preprint, arXiv 2506.08872, 2026.arxiv.org
- Common Sense Education, "Practical Tips for Teachers to Use AI," Feb 2026, source of the personal information rule above.commonsense.org
- Common Sense Education, "How to Help Prevent AI Use for Plagiarism," Mar 2026, source on AI-detector false positives.commonsense.org