Ofsted expects nothing in particular from your school on AI. No policy, no strategy, no named lead, no evidence file. It has said so in writing, and the sentence is blunter than most leaders expect: "We do not expect or require providers to use AI in a certain way, or to use it at all."
That is not the relief it sounds like. Ofsted has also said its inspectors may ask leaders one question, and it is the hardest question about AI anyone will put to you this year.
What does Ofsted expect from schools on AI?
Ofsted expects you to be able to explain your own decisions. Its guidance, How Ofsted looks at AI during inspection and regulation, published on June 27, 2025 and last updated on October 21, 2025, sets no standard for AI use and names no required document. What it names is a question: "Inspectors may ask leaders how they make sure that any use of AI supports the best interests of children and learners."
Read that slowly, because two words in it do the work. How, not whether. And make sure, which asks for a mechanism rather than an intention. A leader who answers "we're being very careful with it" has not answered. A leader who answers "we allow it for these three things, I approved them on these dates, and here is what would make me withdraw that" has.
The guidance applies to providers Ofsted inspects in England, so it does not bind schools in Wales, Scotland or Northern Ireland, or international schools. The question travels anyway. Every inspectorate and every board eventually asks a version of it.
Does Ofsted inspect AI use directly?
No. The guidance is explicit: "We do not look at AI as a stand-alone part of our inspections, and we do not directly evaluate the use of AI."
Ofsted has also been candid about why. It has said it does not yet have the evidence it would need to define good use of AI, which is an unusually honest thing for an inspectorate to publish and worth more to a leadership team than any vendor checklist. Nobody has established what excellent AI practice in a school looks like. Ofsted is not pretending otherwise.
The September 2026 inspection toolkits reinforce the point by omission. Ofsted's own summary of what changed for September 2026, published on June 12, 2026, runs through achievement and context, a widened definition of inclusion, leaders with an inclusion role, learning walks and report card data. AI does not appear in it.
So if you have been sold preparation for an AI inspection, you have been sold something that does not exist. What exists is your ordinary inspection, with AI running through parts of it.
Do schools need an AI policy for Ofsted?
No. Ofsted does not require an AI policy, and a policy on its own would not answer the question it does ask.
This matters because the policy is the thing most schools reach for first, and about half have not got one. The National Education Union's State of Education survey, run online between February 5 and 16, 2026 with 10,311 teachers and 2,996 support staff in English state schools, found that half of teachers report no AI policy at their school: "An average 49 per cent of teachers tell us they have no policy whatsoever on AI, either for staff or students." Two thirds of teachers said there was none specific to student use. Over the same period teacher use climbed from 53 percent to 76 percent. Practice is a long way ahead of paperwork.
A separate study by Accenture and Teach First, published on June 30, 2026 and covering around 200 secondary schools with 30 leader interviews, put the share with a formal AI strategy at 2 percent.
Write the policy if it helps you decide things. Just do not mistake it for the answer. The guidance does note that inspectors may look at what a school has decided about pupils using AI for homework, which is a position rather than a document, and plenty of schools hold that position clearly without having written it down. The reverse is more common: a policy on the server, and no shared position in the building. If you want the distinction drawn out properly, it is the subject of AI strategy versus AI policy.
The four places AI reaches an inspection anyway
AI has no section of its own, so it arrives through four doors that were already there.
Safeguarding. The guidance is direct about this: "AI can pose new and unique safeguarding risks. Inspectors will give appropriate weight to this as part of their more general consideration of a provider's safeguarding culture." Note where the weight falls. Not on an AI safeguarding policy, but on your safeguarding culture, judged as a whole, with AI now among the things it has to hold. That is a higher bar than a document, and it is the subject of what AI means for safeguarding in schools.
Personal data. Inspectors may ask about AI and personal data much as they ask about any other collection, storage and processing of it. The awkward cases are rarely the approved platform. They are the free tool a hard-pressed teacher pasted a class list into, which is why whether staff may put student data into ChatGPT is a question worth settling in plain words before anyone asks it on your behalf.
Bias and discrimination. The guidance warns that AI "will perpetuate bias and discrimination that is present in the data that it processes or has been trained on," and treats that risk as part of what inspectors weigh when collecting evidence. The sharp end is not the lesson resource. It is anything that sorts children: a tool that groups them, flags them, predicts their grades, or drafts a paragraph about them that follows them into the next conversation.
Outcomes and experience. Inspectors may consider the effect of AI on children's outcomes and experiences. Which is the one door nobody has a file for, and the reason the next section exists.
The Best-Interests Answer
The answer an inspector is asking for has four parts, and a leader should be able to give all four standing up. I offer this as a way of preparing rather than a tested method, and it takes its name from Ofsted's own words:
The Best-Interests Answer is what I ask school leaders to be able to give out loud, without reaching for a document, when an inspector asks how they make sure their use of AI supports the best interests of children: name one thing AI now does in this school and who decided it; say what you checked before you allowed it; say how you would know if it were harming a child's learning or safety; and say what would make you stop. A leader who can give all four is describing a school that governs its own use of AI. A leader who reaches for the policy is describing a document.
Four parts, in this order.
- The decision and its owner. One concrete thing AI now does here, and the name of the person who allowed it. "Staff draft parent letters with it, I approved that in September" beats any amount of vision.
- The check before permission. What you looked at before you said yes. Data handling, what the tool does with a child's work, what it blocks, who sees the output. Not all of it. Something.
- The signal it is going wrong. How harm would reach you, and how fast. This is the part almost nobody has, and it is the part that turns care into a mechanism.
- The stopping condition. What you would need to see to withdraw it. A use with no stopping condition is not a decision your school has made. It is a habit it has acquired.
Try it on one use of AI in your school this week. If part three or part four comes out thin, you have found the gap, and you have found it before an inspector did.
The Leadership Question
The question worth asking is not whether you would survive an inspection on AI. It is whether you could say what you are trying to achieve with it.
Ofsted's own research says most schools could not, and the finding is more useful because of who it studied. "The biggest risk is doing nothing": insights from early adopters of artificial intelligence in schools and further education colleges, published on June 27, 2025, is based on 21 interviews with senior leaders and AI leads in the spring term of 2025. These were the keen ones, the schools and colleges furthest along. Ofsted found that they had prioritized safe and ethical use, trained staff before rollout, and in several cases set up approval committees and monthly ethics meetings. Then it found this: "Most relied on feedback from staff and students or tracked and monitored staff usage of AI tools, rather than collecting data that could be used to measure the impact of AI specifically on pupils." And this: "Very few leaders had established what they hoped to achieve with AI longer term, or what success with AI looked like beyond the initial piloting stage."
The best-prepared schools in the country were counting who had logged in.
I spent years on the other side of an inspection, as an assistant headteacher and then as Director of Digital Strategy in a further education college, which Ofsted also inspects. The pattern I saw then holds now. Inspection does not reward the school with the most documents. It rewards the school whose leaders can explain what they chose and why, in their own words, standing up, without notes. Across the leadership teams I work with, the ones who find this easy on AI are not the ones with the longest policy. They are the ones who have said no to something and can tell you what it was.
That is also why who is accountable when AI gets it wrong is the governance question underneath this one. An inspector asking how you make sure is asking, in a politer form, who answers.
What to do this term
Four moves, none of which needs a new document.
Write down the three things AI actually does in your school now, with a name and a rough date against each. Most leadership teams find this list is longer than they thought and less owned than they assumed.
Take the one with the most contact with children and run the Best-Interests Answer over it out loud.
Fix whichever of the four parts came out thinnest. Usually it is the signal, and usually the fix is small: one named person, one route, one sentence to staff about when to raise something.
Then tell your governors, because they will be asked too. What to report to governors about AI covers what that report needs to contain.
If your leadership team wants to know where it actually stands before someone else decides, that is the work I do through school AI readiness reviews. And if you would rather think this through a week at a time, I write about it in my newsletter.
Sources and further reading
- Ofsted, How Ofsted looks at AI during inspection and regulation, GOV.UK, published June 27, 2025, last updated October 21, 2025.
- Ofsted, "The biggest risk is doing nothing": insights from early adopters of artificial intelligence in schools and further education colleges, GOV.UK, June 27, 2025.
- Ofsted, What you need to know about the September 2026 updates to education inspections, Ofsted education blog, June 12, 2026.
- National Education Union, State of education: AI, survey of 10,311 teachers and 2,996 support staff, fieldwork February 5 to 16, 2026.
- Accenture and Teach First, study of around 200 secondary schools, published June 30, 2026, as reported by TechRadar.
Dan Fitzpatrick helps leaders and organizations understand, plan for and lead the changes caused by AI. More about Dan.
