In most of the countries in a new international study, fewer than one third of school principals said their education authority had given them guidance on generative AI. They were not holding back. Between 44 and 85 percent, depending on the country, had already used the tools themselves.
If you are one of those principals, "who will tell me what to do?" is the wrong question. The useful one is "which of my AI questions are mine to settle now?" Sort every question into one of three piles: decide it here, ask someone above you, or record it as open. In my experience the first pile is larger than leaders expect, and the third is the one almost everyone skips.
What does the research say about principals working without guidance?
In most of the countries studied, principals were asked to lead on generative AI without direction from above. The finding comes from an August 2026 brief by Sabine Meinck, Marlen Holtmann, Julian Fraillon and Rolf Strietholt, published by the International Association for the Evaluation of Educational Achievement (IEA) and drawing on ICILS 2023, the International Computer and Information Literacy Study. It reports on more than 1,100 principals in 11 countries, surveyed in late 2023 and early 2024. In most of those countries, fewer than one third of principals reported receiving guidance from educational authorities. In four countries, fewer than 10 percent reported receiving information on explicit policies or recommendations.
Their worries were specific. On average, more than three quarters thought it likely that students would come to rely on the tools instead of learning independently, and around two thirds expected the tools to make deep understanding of a subject harder to build. One concern was shared by almost all principals, whatever their experience: that generative AI would make it harder for teachers to judge whether submitted work is the student's own.
The authors do not read this as reluctance. Their conclusion is that the results "do not suggest resistance to generative AI among school principals," and that principals "cannot be expected to manage this transition on their own."
Two limits apply. The data predates much of the national guidance that exists now, and the 11 countries do not include the United Kingdom or the United States. Treat the study as evidence of a pattern, not a measurement of your system today. You can test whether the pattern holds for you in five minutes: list what your trust, district or authority has sent you on AI, with dates. If the list is short, or older than the tools your staff are using, you are in the pattern.
Why does official guidance leave so much to you?
Guidance written for thousands of schools can mark outer limits, but it cannot settle what only one school knows. England's Department for Education says as much in its policy paper on generative AI in education, first published on October 26, 2023 and dated August 12, 2025 in the version I read: "Schools and colleges are free to make their own choices about the most suitable use cases for generative AI tools in their settings, as long as they comply with their wider statutory obligations such as keeping children safe in education." On homework and unsupervised study, the paper says schools may wish to review their policies, which is an invitation, not an instruction.
Having advised the UK Department for Education, KHDA in Dubai and the Ministry of Education in Kazakhstan, I understand why system guidance reads this way. A document that has to serve every school in a country cannot know your intake, your staff or your tools. That is my interpretation, not the Department's, but the practical result holds wherever you work: the authority marks the edges of the field, and you decide what happens inside it.
The same paper contains one sentence no principal can hand upward: "The quality and content of any final documents remains the responsibility of the professional who produced it and the organization they belong to, regardless of the tools or resources used." Responsibility stays local even when guidance is thin.
How do you sort what is yours to decide?
Ask of each AI question whose decision it is, then put it in one of three piles: decide, ask or record. This is my suggested way of thinking, not an established method, and it works because each pile has a different next move.
| Pile | The test | What you do |
|---|---|---|
| Decide | It affects only practice in your school, can be reversed within a term, and breaks no statutory duty | Decide it, write one sentence, set a review date |
| Ask | It involves children's data, a product another body must approve, an exam board's rules, or a legal duty | Put the question in writing to a named person above you, with the date you need an answer |
| Record | Nobody can settle it yet because the evidence or the rules are still moving | Write down your interim position, your reason and the date you will look again |
The record pile is the one that protects you. When a governor, a parent or an inspector asks what you did about a question nobody answered, "we held this position from this date, pending that decision" is a defensible reply. Silence is not.
What does the sort look like in practice?
Here is a hypothetical secondary school, with five questions staff have actually raised.
- May teachers use AI to draft letters to parents, reading every word before it goes out? Decide. It is local and reversible, and the teacher stays responsible. Review it at half-term.
- May staff paste pupil names and notes into a free chatbot? Ask. Children's data is involved, so the data protection lead, and the trust where there is one, must answer. The interim rule is no. The post on whether staff can put student data into ChatGPT sets out the rule to write.
- Which tool should the school approve? Ask, then follow a process. What schools should require before approving an AI tool covers the checks.
- How should the school handle AI in coursework? Split it. The exam board's rules are fixed, so you ask. How teachers set tasks and check understanding in lessons is yours, so you decide.
- Should Year 7 pupils use AI at all? Record. Age limits on products and the evidence on learning are both moving. The interim position: teacher-led demonstration only, reviewed in January.
Five questions produced one clean decision, two asks, one split and one open record. That is a typical shape, in my experience: far fewer questions need anyone's permission than the waiting suggests.
What should you ask the people above you?
Ask for a decision, not for guidance, because guidance gets you a webinar and a dated question gets you an answer. Write each ask with four parts: the question, why it needs a system answer, what you will do in the meantime, and the date by which you need to know.
Use the study's own three words as a template. It recommends that authorities give "strategic guidance that clarifies expectations, responsibilities, and boundaries." Ask your trust or district: what is expected of us, who is responsible for what, and where are the limits? If nobody can name the person who owns the answer, you have found the real gap, and who owns AI in a multi-academy trust or district is the question to take upward first.
Your governing board sits above you in the line of accountability, and it can often press an authority harder than a single principal can. What a chair of governors should ask the head about AI is a useful page to hand your chair before that conversation.
The mistake I see most often
The commonest mistake is waiting for the permission slip while the staff room writes its own policy. Across the leadership teams I work with, the pattern is a principal who is confident the trust will send something soon, and who therefore decides nothing. Meanwhile individual teachers decide for themselves, one tool and one task at a time. By the time guidance arrives, the principal is reacting to practice that already exists instead of holding a position.
The strongest objection is that deciding before the authority speaks is the riskier course. For the ask pile, it is. Children's data, product approval and statutory duties are hard to take back, so wait for the answer and hold a cautious interim rule. For the decide pile, the risk runs the other way: an unstated position is still a position, only one nobody chose. The working rule is to decide small, reversible things now and ask about anything that is hard to undo.
What should you do this term?
Produce one page, in five steps, within the next two weeks.
- List every AI question staff have put to you since September. Ten minutes with your deputy will surface most of them.
- Sort each into decide, ask or record.
- Write the page: decisions with review dates, asks with names and deadlines, open items with interim positions. Why a short school AI policy beats a long one explains why one page is enough.
- Send the asks this week.
- Tell staff the page exists, and keep any training short. The study recommends "short, modular, and regularly updated professional development," which suits a position you will revise at each review date. The first 90 days of leading AI change shows where this page fits in a longer sequence.
Where to go from here
If your leadership team is working through which AI decisions belong to you and which belong above you, this is the kind of work I support through AI strategy sessions for schools. If you are planning a leadership day or conference session on it, my education keynotes are built around questions like these.
Sources and further reading
- Meinck, S., Holtmann, M., Fraillon, J. and Strietholt, R. "School Principals as Change Agents in an Era of AI-Driven Transformation: Insights from ICILS 2023." IEA Compass: Briefs in Education, No. 31. International Association for the Evaluation of Educational Achievement, August 2026. Publication page and full brief (PDF).
- Department for Education (England). "Generative artificial intelligence (AI) in education." Policy paper, first published October 26, 2023, version dated August 12, 2025. GOV.UK.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and an international keynote speaker who has trained more than 150,000 educators across 30+ countries.
