School leaders should decline five things with AI this year: a pilot with no named owner, a target measured by usage, a detector treated as evidence, a tool that hands students final answers by default, and any move of student data before the rule for it is written. Declining is a decision, not a delay. It is also the part of an AI strategy that most plans leave out.
Why is a list of things to decline part of a strategy?
Because every yes spends the same limited attention, and by October the year's attention is mostly spoken for. Michael Porter put it plainly in his 1996 Harvard Business Review essay What Is Strategy?: "The essence of strategy is choosing what not to do." He added that strategy "renders choices about what not to do as important as choices about what to do."
Now look at what is happening in schools while leaders decide what to do. The RAND Corporation's survey of American schools, AI Use in Schools Is Quickly Increasing but Guidance Lags Behind (September 30, 2025), found that 54% of students and 53% of English, math and science teachers reported using AI for school. Only 45% of principals said their school or district had any AI policy or guidance. The report's own summary is five words: "AI-related training and policies lag."
Read those numbers as a leader would. Use is not waiting for a decision. It is happening in classrooms and staff rooms whether or not anyone has chosen it, so the choices a school does not make get made by default, by the vendor email that arrived on Tuesday or the enthusiastic teacher who signed up for a trial. A decline list is how a leader takes those choices back.
Here is a hypothetical. A principal opens the fall leadership meeting with eleven AI items on the agenda: two pilots, a parent evening, a vendor demo, a staff survey, three policy questions, a grant application and a request from the board. Nobody in the room can say which of the eleven would still matter in June. The strongest move available to that principal is not a twelfth item. It is striking five of the eleven, out loud, with a date to look at them again.
What is the Decline Test?
The Decline Test is a four-question filter for any new AI commitment, and it is my suggested way of thinking, not a researched method. If a proposal fails any one question, it goes on the list of things you decline for now. If it passes all four, it earns a slot.
- Who named the problem? Someone in your building, with a face and a Monday morning, must be able to say what is hard today. A proposal that starts with a capability ("it can summarize anything") and goes looking for a problem fails.
- Who answers when it goes wrong? One named person, not a committee or a vendor, owns the tool, the rule and the outcome. This is the question behind who is accountable when AI gets it wrong.
- How will you know it worked, and by when? There is a date and an observable sign, written down before the start. "Staff feel more confident" is a mood, not a sign.
- Does a human still do the thinking? The student still attempts the problem. The teacher still makes the judgment about the child. The tool takes the typing, sorting and formatting. I call this outsourcing the doing, not the thinking, and it is the test that catches the most damaging mistakes.
Which five things should a school leader decline first?
Start with the five commitments that fail the test most often, because each looks sensible from the inside.
A pilot with no owner and no verdict date
Decline it, because a pilot that nobody has to judge never ends, and it quietly becomes the school's policy. The reasons AI pilots go nowhere are rarely technical. They are failures of Question 2 and Question 3: no owner, no date. If a teacher wants to try something, say yes with a name and a decision date attached. Without them, say not yet.
A usage target
Decline any goal of the form "80% of staff using AI by December." Usage tells you who has opened the tool, not whether anything improved, and it rewards the wrong behavior: people log in to be counted. I have made the case at length that measuring AI maturity means ignoring usage. Replace the target with Question 3.
A detector treated as evidence
Decline the use of an AI detector as the proof in a misconduct case. My view, which I hold firmly, is that detectors are unreliable enough to be harmful in a disciplinary setting, and I have set out what to put in your policy instead. The human stakes are visible in the RAND data: half of students said they worry about being falsely accused of using AI to cheat. A school that treats a score as a verdict teaches children to fear the tool and conceal their use of it. When a student is accused, the process matters more than the software.
A tool that hands students the answer by default
Decline any product that produces the finished solution before the student has tried. England's Department for Education now says this in terms. Its Generative AI: product safety standards, published January 19, 2026, expect developers to "make every effort to mitigate the potential for cognitive deskilling," and "not to provide final answers, full solutions, or complete worked examples by default," with a full solution shown "only after a genuine learner attempt." The standards are written for suppliers, and the Department notes that schools "may also find them helpful in assessing which AI products are safe for use in education." They give you a ready-made question for any vendor conversation. This is Question 4 in the Department's own words. It also pairs with the checks in what to require before approving an AI tool.
Moving student data before the rule exists
Decline every workflow that touches student information until the rule for it is written and staff have read it. The sequence matters. A rule that arrives after the habit is a correction, and corrections are slow. The rule to write about student data and ChatGPT is short, and it is the easiest decision on this list to make early.
What do I tell leadership teams about declining?
The pattern I see most often is that leaders hear "decline" as "fall behind." In my advisory work with government education bodies, the plans that hold up are the ones that name what they are leaving alone and why, because a plan that leaves everything open gets stretched until it fits nothing. The conversation usually changes when I move the question from "what if we miss something?" to "what are we already doing badly because we said yes to it?"
The honest counterargument is that a decline list can turn into a stall. It does when the declines have no dates. "Not now" with no revisit date is just "no" without the courage. Every item on the list therefore needs an owner and a day it comes back. The leaders I see doing this well treat the list as a living document, and they read it to their staff, which does something a strategy document rarely does: it tells people what they may stop worrying about. If you are unsure where you stand before you start declining, are we behind on AI? is the place to calibrate.
How do you write a Not-Now Register?
A Not-Now Register is a one-page table with four columns: what you declined, which test it failed, when it comes back, and who brings it back. Copy the structure below and fill it with your own items. The two example rows are hypothetical.
| Declined for now | Test failed | Comes back | Owner |
|---|---|---|---|
| School-wide AI feedback tool for essays (hypothetical) | Q3: no sign of success defined | Mid-year review, after a one-class trial with a written measure | Head of English |
| Student-facing chatbot in the library (hypothetical) | Q4: gives answers before an attempt | When the supplier can show the answer-after-attempt setting | Deputy head, curriculum |
| Your item |
Use it in three steps. First, run each item on your next agenda through the four questions and put the failures in the table. Second, give each failure an owner and a date, the first checkpoint no later than the next mid-year review. Third, share the register with staff and with your governors or board, so that a decline is a visible decision and not an unexplained silence.
What should you do next?
Take your own agenda this week, count the AI items on it, and decline the ones that fail the test. A short list you can defend in a board meeting is worth more than a long one nobody can finish. If your leadership team is working through these choices, this is the kind of work I support through AI strategy sessions and advisory work for schools.
Sources and further reading
- Michael E. Porter, What Is Strategy?, Harvard Business Review, November-December 1996.
- Christopher Joseph Doss and colleagues, AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels, RAND Corporation, September 30, 2025. Nationally representative US samples; findings are American and from spring 2025.
- Department for Education (England), Generative AI: product safety standards, GOV.UK, published and updated January 19, 2026. Written for England and for suppliers; schools may use it to assess products.
Dan Fitzpatrick is the founder of The AI Educator and advises schools, districts and education systems on leading through AI. About Dan.
