An AI Framework for Schools and Teachers: A Practical Guide to Getting It Right

1. Where schools are now: students ahead, teachers without guidance

Generative AI reached most classrooms on students' phones before any school policy existed. Surveys in 2024 and 2025 found the same pattern across countries: most secondary students had used a chatbot for schoolwork, and only a minority of teachers said their school had given them clear guidance.

Schools tend to respond in one of two ways. A ban cannot be enforced, and students keep using the tools at home without telling anyone. An open door with no structure leads to copied work and less practice at the thinking school is meant to build. Both leave teachers without direction.

The same research shows what teachers gain when they use the tools deliberately. Drafting a lesson plan, adapting a worksheet for three reading levels, writing first-pass report comments and building a quiz all take less time. The framework below keeps that gain and closes off the failure modes. It rests on one rule: AI belongs wherever it protects students' thinking and nowhere it replaces it.

2. Five pillars to check when something goes wrong

The framework has five pillars, short enough for a busy vice-principal to remember. They work at the same time, not in sequence, and each one gives you a separate place to look when a problem appears.

Literacy. Everyone in the building, staff first and then students, learns what these tools do and where they fail. A chatbot predicts likely text; it does not look facts up unless it is connected to search. It invents citations. It repeats the biases in its training data. Staff who know these limits can supervise the tool.

Practice. The approved ways AI appears in lessons. A teacher uses it to plan and differentiate. A student uses it for feedback on a draft, to generate practice questions or to argue against their own essay. This pillar needs worked examples, which section 4 provides.

Integrity. The rules and, above all, the assessment design that keep AI from doing the student's thinking. Set tasks where outsourcing the work defeats the purpose of the task, and you catch fewer cheats because there are fewer to catch.

Privacy. No child's name, image, grades or personal circumstances go into a public AI tool. This pillar carries legal weight in most countries, and staff breach it by accident more often than on purpose.

Equity. If AI help depends on having a phone, a data plan and a quiet room at home, the school widens the gap it exists to close. The fix is supervised access on school devices during school time.

3. A six-stage rollout you can start this term

Schools that manage this well do not switch AI on across the building in September. This is the sequence I recommend to the schools I work with.

1. Form a working group of four to six people. Include a senior leader, two or three teachers who are curious, someone who handles data or IT and, if possible, a parent. Choose the most trusted staff, not the most technical.
2. Write a one-page position. Before choosing any tool, agree in plain language what AI is for at this school, what it is not for and who to ask. Keep it to one page so parents and staff read it.
3. Train staff before students. Run two or three short hands-on sessions where teachers use the tools on their own tasks: planning a lesson, marking a sample, building a quiz.
4. Pilot in two or three classrooms. Pick willing teachers in different subjects. Give them approved uses, a simple log of what worked and permission for some lessons to fail. Run it for one term.
5. Redesign the assessments that break. Use the pilot to find which tasks a chatbot can now complete in seconds, and change those tasks.
6. Publish, widen and review. Share what the pilot found, extend approved uses to more classes and set a review date. The tools change every few months, so review at least once a year.

4. Five classroom moves to try on Monday

These are moves I have seen work in real lessons.

Plan and differentiate. Ask the tool to turn one lesson objective into three versions of a task at different reading levels, then edit them. Teachers tell me this is where they save the most time.

Make the AI the student. Show the class a chatbot's answer and ask: where is it wrong, what did it leave out, what sounds confident but has no support? Students learn to doubt fluent text, and the tool becomes something they study instead of something they copy.

Assess the process. Let students get AI feedback on a draft, then hand in the draft, the feedback and their revision together. You grade how they responded to feedback, which a chatbot cannot do for them.

Socratic practice. Set the tool to ask questions instead of giving answers, so a student stuck on a maths problem gets a hint and not the solution. Several school-focused tools now work this way by default.

Cite AI like a source. Each time students use AI, they note where and how. This makes honest use normal and keeps the tool visible.

The tool below shows how teachers describe the change in their week. Switch between the two views: the total hours stay the same, and more of them go to work only a teacher can do.

5. Guardrails: integrity, privacy, equity

Integrity without detectors. AI detectors miss edited AI text and falsely accuse students who wrote their own work, and the false accusations fall hardest on students writing in a second language. Protect integrity through task design instead: more writing done in class, short oral defences of submitted work, drafts and notes handed in with the final piece, and assignments that ask students to correct or challenge AI output. When cheating takes as much effort as the task, most students do the task.

Privacy as a fixed rule. No student's personal data goes into a public AI tool: no name attached to a grade, no photo, no description of a child's home situation for a report. Staff who want AI help with sensitive work either remove identifying details first or use a school-approved tool covered by a data agreement. Print this rule in bold on the one-page policy.

Equity through provision. Give every student supervised access on school devices and in timetabled lessons, so a student without a phone or home internet still builds the skill. That means a budget line for devices and connectivity and a slot on the timetable.

6. Measure your school's readiness

The tool below is a working version of the five-pillar model. Set your level on each pillar and the chart shows which one is holding the others back. There is no pass mark; use the result to decide where the working group spends its first term.

7. The mistake to avoid, and the one to make

The mistake to avoid is treating AI as an IT project. Schools that bought a tool and expected it to deliver the framework have struggled. Schools that treated it as a change in teaching practice, supported their teachers and accepted an untidy first term have done better.

The mistake worth making is starting before you feel ready. The tools change faster than any policy cycle, and while a working group waits, students keep using AI with no guidance. A pilot in three classrooms this term will teach you more than another year of meetings.

The part of this framework I am least sure of is equity. Supervised access on school devices is right in principle, but many Caribbean schools share a single computer lab between hundreds of students, and I have not yet seen a low-cost model that gives every student regular time on a capable tool.

"The schools that get this right are not the ones with the biggest budgets. They are the ones that trained their teachers first and trusted them to lead." - Adrian Dunkley, AI Boss

8. The published frameworks this draws on

The five pillars combine themes that national and international guidance keeps returning to: clear policy, AI literacy, data privacy and phased introduction. These are the primary sources worth reading.

• AILit Framework, a joint European Commission and OECD initiative supported by Code.org, defines what learners should know and be able to do in the age of AI. ailiteracyframework.org
• TeachAI: AI Guidance for Schools Toolkit, ready-to-adapt policy language organised as cycles of review. teachai.org
• Australian Framework for Generative AI in Schools, six principles covering teaching, wellbeing, transparency, fairness, accountability and privacy. education.gov.au
• The SEE Framework from AI for Education, a practical route to building generative AI literacy. aiforeducation.io
• Independent Schools NSW phased implementation framework, and the NCEE Framework for AI-Powered Learning Environments for system leaders.
• UNESCO guidance on generative AI in education, which recommends a minimum age of 13 for independent use.

9. What to do next

1. Principal: name the four-to-six-person working group this month and give it a deadline of six weeks for the one-page position.
2. Working group: draft the one-page position using the six items in the FAQ below, and send it to staff and parents before any student uses a tool.
3. Deputy or head of professional development: book two hands-on staff sessions of about 90 minutes before the pilot starts, with each teacher bringing one real lesson to plan with AI.
4. Heads of department: list every take-home assessment for the coming term and test each one in a free chatbot. Any task it completes well in under a minute gets redesigned before it is set.
5. Data or IT lead: check the terms of each tool staff want to use for age limits and where data is stored, and publish a short approved list.

Frequently asked questions

What age is appropriate for students to use generative AI?

Thirteen is the usual floor for independent use: most major services set it as their minimum age, and UNESCO recommended it in its 2023 guidance on generative AI in education. Check the terms of the specific tool, because some vendors offer education accounts with different age rules and data terms from their consumer apps. Students under 13 can still learn about AI through teacher-led demonstrations on the teacher's account.

Are AI detectors worth paying for?

Not as evidence for discipline. A 2023 Stanford study led by Weixin Liang found that popular GPT detectors flagged more than half of essays written by non-native English speakers as AI-generated, while scoring essays by native-speaking students accurately. Detector vendors themselves advise against using a score as the sole basis for an accusation. If the school already has one, treat a high score as a reason to talk to the student about their drafts, nothing more.

What should go in a one-page school AI policy?

Six short items cover most schools. State what AI may be used for, what it may not be used for and which tools are approved. Then add the rule that no student personal data goes into a public tool, how students note their AI use, and who to ask. Add the date it will next be reviewed. Leave detailed procedures to separate staff guidance so the page stays readable for parents.

Which law covers student data in a Jamaican school?

Jamaica's Data Protection Act 2020, which came fully into force in December 2023, applies to personal data a school collects, including grades, photos and records about a child's circumstances. The Office of the Information Commissioner oversees it and publishes guidance for organisations. Schools in other countries should check their national data protection authority, since most have similar rules on children's data and on sending data to overseas services.

How should a school explain its AI approach to parents?

Send the one-page policy home before students start using any tool, with one worked example of permitted use and one of misuse. A 30-minute evening where a teacher shows a real lesson that uses AI answers more questions than a letter. Name a staff member parents can contact, and tell them when the policy will be reviewed so they know it is not fixed.