The most useful question a governor can ask about AI is also the hardest to answer: when did we last say no?
A head who can answer it has a gate. A head who cannot has either a gate nobody uses or no gate at all, and either way the board has just learned something no policy document would have told it.
Most AI questions in governing body meetings work the other way. Do we have a policy? Have staff been trained? Are we keeping this under review? Each takes one word to answer, the item moves on, and the minutes record that governance took place.
What should a chair of governors ask a headteacher about AI?
Ask six questions, and phrase each one as "when did we last" instead of "do we have", because a "do we have" question can always be answered with a yes, while a "when did we last" question can only be answered with a story or an honest silence.
I call this the Last Time Rule. The Last Time Rule is the habit I teach chairs: never ask whether the school has something; ask for the last time it was used. "Do we have an approved tools list?" earns a yes. "When did a tool last come off the list, and why?" earns a date, a reason, or a pause, and the pause is the finding. This is my suggested way of questioning, not a validated method. It needs only a chair willing to wait through the pause.
The questions are written for England's governing boards and travel elsewhere. In a US district, read "board president" for chair and "superintendent" for headteacher; the questions work unchanged.
Why does "do we have a policy?" tell a board nothing?
Because it asks about a document, and a board's work is to oversee decisions. As The Key summarizes the Department for Education's maintained schools governance guide (published November 19, 2025, updated July 15, 2026), a governing board has three core functions: clarity of vision, ethos and strategic direction; holding executive leaders to account; and overseeing financial performance. The second is the one these questions serve. A policy is evidence that someone wrote something. It says nothing about whether anyone is deciding anything.
Inspectors read it the same way. Ofsted's statement on how it looks at AI, published on June 27, 2025 and linked from Ofsted's approach to AI, says inspectors assess the provider's decision-making and its effect on children, not the tool itself, and that they may ask how leaders make sure AI use supports children's best interests. Ofsted also says the statement is not a final position and will change as understanding develops. A board that has never asked its head that question has no way to know what the answer would sound like.
The Department for Education's generative AI product safety expectations, first published on January 22, 2025 and updated on January 19, 2026 to add standards on cognitive development, emotional and social development, mental health and manipulation, are addressed to suppliers. Schools may use them to judge products. Nothing on the page tells a board whether its school did. Only a question finds that out.
The six "last time" questions
Each question below asks the head for a specimen: a date, a decision, a person. What a good answer sounds like matters less than whether one arrives at all.
1. When did a member of staff last ask to use a new AI tool, and what happened to the request?
This question finds out whether an approval route exists in practice or only on paper. A good answer has a rough date, a role (not a name), a decision and a reason. A weak answer is "staff know to check with me", or "it hasn't come up". If it has not come up, look at the wider picture. In a Morning Consult survey for IBM of 1,019 US K-12 education professionals, fieldwork in July 2026, 76 percent of middle school and 73 percent of high school educators reported using AI in the classroom weekly. A school in that picture where nobody has asked is a school where staff are deciding for themselves, which is the pattern I describe in what leaders should do about shadow AI.
2. Pick one tool staff use. Where does the data typed into it go, and who could tell me without looking it up?
The second half is the real question. A good answer names one person, usually the data protection officer, who can say where the data goes and has a record of having checked. A weak answer is "it's GDPR compliant", which is a supplier's claim repeated. The checks behind a good answer are in what schools should require before approving an AI tool and whether staff can put student data into ChatGPT.
3. When did you last learn something about how staff or pupils use AI that surprised you?
A head who has looked finds surprises. A good answer names one, says how the school found it, and says what changed afterward. A weak answer is "no concerns", which describes either a school where nothing is happening or a school that is not looking. Asking staff directly, and in a way that makes honest answers safe, is the subject of how a principal can survey staff about AI.
4. When did we last say no?
A school that has used AI for a year and refused nothing has an approval step that approves. A good answer is a request declined, or an approved tool withdrawn, with the reason. Tools change under a school's feet, so the withdrawal case matters as much as the refusal: see what to do when an approved AI tool changes. A weak answer is a proud "we've had no problems", which is an answer to a different question.
5. When did you last test the answer you would give an inspector?
Ofsted's statement says inspectors may ask how leaders make sure AI use supports children's best interests. Ask the head to give that answer now, aloud, as if the board were the inspector. A good answer is short, specific and consistent with the answers to questions 1 to 4. A weak answer starts with the policy. What Ofsted does and does not expect is covered in what Ofsted expects from schools on AI.
6. When did the board last use AI for its own business, and under whose rule?
A board that exempts itself teaches the school that the rule is optional. The National Governance Association's February 5, 2026 post, Keeping it human: Governance professionals and AI, advises using tools provided or endorsed by the school or trust's IT team or data protection officer, and says a data protection impact assessment is needed when a third party will handle personal or private data such as board papers or trustee contact details. If clerk, chair or governors are pasting board papers into a chatbot, the board is the first place the answer to question 2 is wrong.
What I tell chairs before the meeting
I tell chairs to ask no more than two of these at a meeting, to tell the head which two a week ahead, and to ask for the answer to be minuted.
Across the governing boards I work with, the commonest failure is not ignorance but politeness. Governors are volunteers, the head is a full-time professional, and the AI item comes at the end of a long agenda. So they nod. The Last Time Rule gives the chair something polite to say that cannot be answered politely. Jim Batson, a 23-year school board member writing in the May/June 2026 Illinois Association of School Boards Journal, puts the spirit right: board questions should seek understanding, not critique the administration's work. A head who knows the question in advance and is asked it in that spirit will bring a real answer. A head who is ambushed will bring a defense.
I have advised the Department for Education, and the two AI documents cited above show a pattern I would want any chair to notice: Ofsted's statement describes what inspectors do, and the product safety expectations are written for suppliers. Neither is addressed to the person who has to ask the head the next question.
What does question 4 sound like in the room?
Here is a hypothetical, which I have made up to show the technique and which describes no real school.
The chair asks: "When did we last say no to an AI tool?" The head answers: "We haven't had to. Staff are sensible." The chair does not move on. "Is that because no one has asked, or because no one could be refused?" There is a pause. The head says, "Honestly, I'm not sure anyone would ask." That sentence is the finding. It goes in the minutes, and it becomes next term's agenda item: who does decide, and how would a member of staff know?
Notice what the chair did not do. The chair did not say that the head was wrong, and did not name a tool. The chair asked for the mechanism.
Where does asking become running the school?
Governors ask about mechanisms, not products. The board may ask what happens when a member of staff wants a new tool, who decides, and how the school would know if it went wrong. It should not ask which tool, which prompt or which lesson. The test is whether the head's answer could be checked by someone outside the building: a person, a date, a record. If it could, the board is overseeing. If the question requires the board to hold a view on the tool itself, it has crossed into running the school.
This is the counterargument to everything above, and it is right in one case. If the head's answers to questions 1 to 4 are good, the board should stop asking and trust them. The Last Time Rule is for finding out whether the answer exists. Once it does, it earns the head the right to be believed.
A year of questions
A termly board can cover all six in a year without one meeting feeling like an audit. As a hypothetical pattern: questions 1 and 2 in the autumn, 3 and 4 in the spring, 5 and 6 in the summer, with the head's answers carried forward so each term's questions test the last. By the third meeting, the chair is no longer asking whether the school has a handle on AI. The minutes already show.
If your board is working out what it should be able to ask, and your head what it should be able to show, this is the kind of work I support through AI readiness work with school leadership teams and board and conference sessions through education keynotes. New questions like these go out in my newsletter.
Sources and further reading
- Department for Education, Maintained schools: governance guide, GOV.UK, published November 19, 2025, updated July 15, 2026.
- The Key for School Governors, The 3 core functions for maintained boards, summarizing the DfE guide.
- Ofsted, Ofsted's approach to AI, GOV.UK, updated October 21, 2025; includes the statement "How Ofsted looks at AI (artificial intelligence) during inspection and regulation", June 27, 2025.
- Department for Education, Generative AI: product safety expectations, GOV.UK, first published January 22, 2025, updated January 19, 2026.
- National Governance Association, Keeping it human: Governance professionals and AI, February 5, 2026.
- Jim Batson, A Board Member's Guide to AI, Illinois Association of School Boards Journal, May/June 2026.
- IBM and Morning Consult, study of US K-12 education professionals and parents on AI adoption and readiness, September 2, 2026.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and an international keynote speaker who helps school leaders and boards understand, plan for and lead the changes AI brings.
