A new AI lead should do one thing before touching a tool: get a four-line mandate agreed in writing with the head. What you may decide alone, what you must take to someone else, when and how you report, and what is not yours. Then run the first term in order: audit, one page of guidance, one visible decision, one report.
A few weeks into the school year, a head or principal asks someone to "take on AI." The person says yes. They get a title, half a sentence in the staff bulletin and a calendar that was already full. Nobody says what they are allowed to decide.
Then the messages start. Can I use this for reports? Is that tool allowed? A parent has asked about homework. The new lead answers every one of them, and by half-term they are the school's help desk with a better job title.
Without the mandate, the first term's work has nobody standing behind it. With it, every later request lands on authority the lead has already secured.
What should a new AI lead do first?
A new AI lead should ask for the mandate before the first staff meeting, because every later request lands on whatever authority they have already secured. I call the four lines the Mandate Check, and I offer it as a way of thinking, not a tested method:
The Mandate Check is four lines I ask a new AI lead to get agreed in writing before the first staff meeting: What may I decide on my own? What must I take to someone else, and to whom? When do I report, and in what form? And what is explicitly not mine? A lead with all four has been given a job. A lead with fewer has been given a title.
It is the other half of an exchange I wrote about in who should own AI strategy. There I put three questions to the people doing the appointing, and I call them the Ownership Test:
The Ownership Test is three questions I ask before a leadership team appoints anyone to lead AI: Which of the three jobs are we handing over, the decision, the delivery or the expertise? What has come off that person's plate to make room? And when this goes wrong, whose name is on it? A team that can answer all three has delegated. A team that cannot has appointed someone to be blamed.
The Mandate Check asks the same things from the other chair. If your head never ran the Ownership Test, you can run it for them, politely, in one email.
Why does the title arrive without the authority?
The title arrives without the authority because a title asks nothing of the head, while authority asks them to give up a decision they would otherwise have made. That is my reading, and it is easy to miss from the inside, because the head usually means well.
There is some evidence on how schools staff this work. In research published on GOV.UK on June 27, 2025, Ofsted interviewed leaders at 21 schools, multi-academy trusts and colleges in England that had started using AI early. All of them described senior leaders backing AI at a strategic level, usually alongside an AI champion. One trust digital lead put the requirement plainly: "What you really need is someone with leadership responsibility." In smaller providers, the report says, one person often held both the strategic and the champion role. Standalone schools found it harder to draw on data, IT and curriculum leadership together and often used staff working groups instead.
Read that with its limits in view. The study was small and qualitative, and it spoke only to leaders, not to teachers or pupils, so it describes how early adopters organized themselves rather than showing that any arrangement worked. But it names the trap precisely. When one person is both the strategy and the enthusiasm, they carry the exposure of both and only the authority the head has actually handed over. Why AI champions programs fail describes where that ceiling sits: the point at which the next step needs a decision rather than a demonstration.
What does a good mandate look like?
A good mandate is four plain sentences, specific enough that a third person could tell whether you stayed inside them. Here is one filled in for a hypothetical secondary school. The detail is invented to show the shape; yours will differ.
| Line | Weak version | Strong version (hypothetical) |
|---|---|---|
| What I may decide alone | "Lead on AI." | Which uses staff may try for planning and feedback, using tools already on the approved list, and the wording of the staff guidance page once the head has read it. |
| What I must take to someone else | "Anything important." | New tools go to the head, with the data protection officer's view and the designated safeguarding lead's view attached. Anything involving pupils' personal data goes to the data protection officer first. |
| When and how I report | "Keep SLT updated." | One page to the leadership team each half-term, under the same four headings, and one page for the governing body each term. |
| What is not mine | Left blank | Safeguarding decisions, the decision to restrict pupil use, and any staffing consequence of a tool. Those stay with the safeguarding lead and the head. |
The fourth line does most of the protecting. If an AI tool goes wrong, the question that follows is whose decision it was, and a lead who wrote down what was never theirs has an answer. It also keeps the role honest, because a lead who quietly absorbs safeguarding and procurement ends up accountable for work nobody agreed to give them.
The ask can be short: "Before I start, can we agree four things in writing: what I can decide alone, what I take to you, when I report, and what stays with you?" Most heads agree within a day. The ones who hesitate have told you something useful about the term ahead.
What should the first term look like?
The first term should run four jobs in a fixed order: audit, guidance, decision, report. The order matters more than the dates, which are my suggestion for a twelve-week term and not a tested schedule.
| Weeks | Job | What you finish with |
|---|---|---|
| 1 to 2 | Agree the mandate | The four lines, in an email the head has replied to |
| 2 to 5 | Audit what is already happening | A list of the jobs people are using AI for, with the tools and accounts involved |
| 5 to 7 | Write one page of guidance | The page below, signed off by the head |
| 8 to 10 | Make one visible decision | One tool approved or removed, with a name and a date |
| 11 to 12 | Report | The leadership team's one-page report, with what you need next term |
The audit. The Department for Education's support for school and college leaders, last updated on May 19, 2026, includes material on how to audit your current use of AI, and says that any use needs to be carefully planned. Ask the audit as a question about jobs, not a confession about tools: what are you using AI for, and what would you do if you could not? You are looking for the gap between what staff do and what the school has said.
The guidance. Guidance is where a new lead earns the right to stop answering messages. In a Gallup survey with the Walton Family Foundation, published on May 26, 2026, 82 percent of US public K-12 teachers said they had received no formal guidance on using AI at work. Another 48 percent had only informal guidance, such as verbal advice or shared norms. The survey covered 2,069 teachers and was fielded between February 9 and March 2, 2026. It is US data, and I could not find an equivalent figure for England, so treat it as a picture of the problem, not a measurement of yours. Gallup also found that teachers with formal guidance were more likely to say it encourages AI use: for making student materials, 69 percent against 51 percent with informal guidance. Written guidance does not only restrict. It tends to give people confidence.
One page is enough. Four headings:
- You may. The uses the school has agreed, in the staff's own words.
- You may not put in. The data that never goes into a tool, stated plainly enough to recite.
- Ask first. The uses that need a conversation, and with whom.
- If it goes wrong, tell. One name and one route.
The decision. Pick one tool and settle it: approve it or take it off. What schools should require before approving an AI tool gives the one-page record. A single decision that is visibly made, with a name and a date, does more for a new lead's standing than a term of workshops.
The report. Four lines to the leadership team and, a term later, to the board. What to report to governors about AI shows the shape. The last line of your first report should be the request: the decision you need next term.
What I See in Practice
The mistake I see most often is the AI lead who becomes the help desk. Across the leadership teams I work with, the pattern is consistent: the people asked to lead AI are usually the keenest, so they say yes to everything, and every answer they give is a decision that was never made by anyone with the authority to make it. The same question arrives from six people. It gets six slightly different answers. By half-term the school has six policies, all of them in the lead's head.
The fix is dull and it works. Answer each question once, in the guidance page, and send people there. When the same question comes back a third time, it is not a staff problem. It is a missing line on the page. I say this as a former assistant headteacher and now a school trustee, so I have been both the person asking a colleague to take something on and the person being asked. The requests that land well are usually the ones that arrive with a decision attached.
When is an AI lead the wrong answer?
An AI lead is the wrong answer when the head will not give the role any decision rights, because a title without them just moves the exposure onto one person. The counterargument deserves a fair hearing: in a small school, a full mandate can feel like ceremony. One teacher, one head, one Friday conversation.
The scale can shrink; the four lines cannot. In a very small school the mandate might be four sentences in a shared note. If the head genuinely wants to keep every decision, that is a legitimate choice, but call the role what it is, an AI coordinator, and be honest about what it can deliver. AI in a small primary school with no IT team sets out what the head decides in that setting. And if you are the head, the first 90 days of leading AI change sets out where this appointment sits among the other jobs arriving in the same fortnight.
What should you do this week?
This week, send the four lines to your head and ask for a reply. Then put the first report in the diary for twelve weeks out, so that the deadline exists before the work does. Do not open a single tool until both are done. If you have already been answering questions for a month, start at the guidance page instead: write the one page from the questions you have been asked, and send people to it.
If your leadership team is appointing an AI lead, or has already done so and wants the role to hold, this is the kind of work I support through leadership mentoring and AI strategy sessions.
Sources and further reading
- 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. Read it
- Gallup and the Walton Family Foundation, "Most Teachers Receive No Formal Guidance on AI Use," May 26, 2026 (2,069 US public K-12 teachers; fielded February 9 to March 2, 2026). Read it
- Department for Education, "Using AI in education: support for school and college leaders," GOV.UK, first published June 10, 2025, last updated May 19, 2026. Read it
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. More about Dan.
