The teacher at the back of your September AI session, arms folded, is the most useful person in the room. In the United States she is now in the majority. EdChoice's Spring 2026 Teacher Survey, run by Morning Consult with 1,031 teachers in April 2026 and published on 7 May, found that 55% of teachers oppose the use of AI programmes in the classroom and only 38% support it, a fall of eight points in support since the autumn. Sixty-five per cent oppose pupils using AI for schoolwork. A year of pilots, vendor demonstrations and training days has produced more sceptics, not fewer.
Most leadership teams read that as a persuasion problem. It is not.
You bring the AI sceptics with you by finding out what each of them is actually objecting to, providing it, and giving the ones with a principled objection a real job in the decision. You do not do it by persuading them, by putting the enthusiasts in front of them, or by booking another training session. In the adoption work I do with leadership teams, the sceptics who came across did so when the organisation changed, not when their minds did.
Why the sceptics are multiplying, and why they are often right
The sceptics are multiplying because they are being asked to use a tool without a reason, a rule, a result or a safety net, and the evidence says that is the ordinary condition of teaching with AI in 2026. Gallup's Teaching for Tomorrow study for the Walton Family Foundation, published 26 May 2026 from 2,069 US teachers surveyed in February and March, found that 18% had received any formal guidance from their school on how AI should be used, 34% had received none at all, and for one-to-one tutoring the no-guidance figure was 69%. Where guidance existed, most teachers said it neither encouraged nor discouraged use. That is not a workforce refusing to change. It is a workforce left to decide alone and deciding, sensibly, not to.
The pattern in England is the same shape at a smaller scale. Teacher Tapp's April 2026 data shows AI use for lesson planning rising from 14% to 20% between December and April, but with maths teachers at 6%, and one teacher's worry recorded beside the figures: that newer colleagues are losing the thinking time that designing a lesson from scratch used to force on them. That is a sceptic making an argument about learning, not about software.
And sometimes the sceptic is simply correct about the tool. In METR's randomised trial of July 2025, sixteen experienced open-source developers were timed across 246 real tasks with and without AI coding tools. With AI they took 19% longer. They had predicted a 24% speed-up beforehand, and afterwards still believed they had been sped up by 20%. The authors are careful to say the result applies to experienced people working in familiar territory and should not be stretched to every domain. I would add: experienced people working in familiar territory is a fair description of most of the sceptics in your staffroom. The people who trust the tool most check it least, and the person who refuses to believe the time saving until she has seen it is doing the checking for you.
Those are the findings. What follows is my interpretation, and what I tell leadership teams to do with it.
What is a sceptic actually objecting to?
A sceptic is not someone who says no to AI. A sceptic is someone who has not yet been given a reason, a rule, a result or a safety net, and who is honest enough to say so. Persuasion fails because it argues with the wrong thing: it makes the case for the tool when the objection is about the organisation.
The Sceptic's Question is the one question I ask leaders to put to every member of staff who is holding back from AI, before anyone tries to persuade them: what would have to be true for you to use this on your own work? The answer names what the organisation has not yet provided. It is almost always one of four things: a purpose, a permission, a proof or a protection. Each is a leadership job, not a training need.
This is my suggested way of thinking about it, built from the conversations I have after keynotes and in adoption work, not a validated instrument. The World Economic Forum's five faces of AI readiness (June 2026) sorts staff into enthusiasts, curious, cautious, sceptics and opposed. Personas tell you who is in the room. The Sceptic's Question tells you what to do on Monday. Here are the four answers, what each sounds like, and what it asks of you.
A purpose. It sounds like: "What is this for?" or "We have enough initiatives." This sceptic has noticed that AI has arrived attached to nothing, and that nobody has said which of the school's existing problems it is meant to move. AI attached to nothing changes nothing, and this person is refusing to pretend otherwise. What it asks of you: name the one problem this year that AI is for here, out loud, and be able to say what will be different by Easter. If the leadership team cannot agree on that sentence, the sceptic has found your real problem, and it is not her.
A permission. It sounds like: "Am I allowed?" or "What about the data?" or, most often, silence. This is the largest group and the least visible, because a person who does not know the rules does not raise her hand; she waits. Gallup's 18% formal guidance figure is the size of this group in the US. What it asks of you: a short written answer to "what may I use AI for, what may I never put into it, and who do I tell when it goes wrong", read aloud by the head rather than uploaded to the shared drive. That is the permission condition in how leaders build AI confidence across a workforce, and it takes a morning.
A proof. It sounds like: "Show me it works here, with our pupils, on my marking." This sceptic is the one who read METR, or would have. She has watched enthusiasts report time savings that never appeared in her own week. What it asks of you: a result from your own building, measured, with her inside the measurement. Put her in the pilot, give it a decision date, and let her be the one who reports. A pilot run entirely by volunteers proves nothing to anyone who was not in it, which is why AI pilots so often go nowhere. Proof owned by a sceptic travels; proof owned by a champion is discounted on arrival.
A protection. It sounds like: "What happens if it gets something wrong?" or "I am not putting my name on that." This sceptic is asking whether it is safe to be seen using AI and safe to admit a mistake, and she is right to ask, because in most organisations the answer has never been given. What it asks of you: a rehearsed route for the first error, said in public, with the senior leader's own AI mistake told first. That is the fourth question of the Sponsorship Test: have we made it safe to be seen using it? An organisation that treats the first AI mistake as a disciplinary matter has just told its sceptics they were right.
None of the four is fixed by a better slide deck. None of them is a gap in the sceptic. All four are gaps in the organisation, which is why the sceptics multiply after a training day: the day supplies skill, and the sceptics were never short of skill.
The fifth answer: when the sceptic is right
The fifth answer to the Sceptic's Question is "nothing", and it is the most valuable answer you will hear. This is the principled objector: the teacher who believes that a particular use of AI damages learning, or the relationship between a teacher and a child, or the child's own effort, and who would not use it however well it worked. Anne Lutz Fernandez makes that case at length in Resisting AI Mania in Schools (Rethinking Schools, winter 2025-26), and her sharpest point is aimed at leaders: calling scepticism "fear" is a way of not having to answer it.
She is right about that. Call it fear and you have stopped listening. Teachers get labelled resistant to change far more often than they are; most of the time they need time and space, and given both they become the people who drive the change. The principled objector needs something more specific. She needs a job.
The job is the list of what your organisation will not use AI for. Every organisation that is serious about AI has one, said in public, and it is the seventh of the behaviours in what responsible AI adoption looks like in practice. Nobody writes that list better than the person who has thought hardest about the harms. Give her the pen. Ask her to draft the sentences that begin "We will not use AI to", and to argue each one in front of the leadership team. Some of her lines will survive. Some will be narrowed to "not yet, until we have seen". Either way she has moved from the back of the room to the table, without changing her mind, and the organisation has a better policy than the enthusiasts would have written.
You do not convert sceptics. You answer four of them and you employ the fifth.
The Mistake I See Most Often
The mistake I see most often is sending the sceptics on more training. It is the natural move: the enthusiasts loved the session, so the sceptics must need a second one. What actually happens is that the sceptic sits through a demonstration of a tool she has already decided not to use, for reasons nobody has asked about, and leaves with her scepticism confirmed and a new grievance about the afternoon she lost. Training answers "how". The sceptics are asking "why", "may I", "does it" and "what if".
The second mistake is counting the wrong people. Leadership teams report adoption as the share of staff who have used a tool, which is the enthusiasts plus the compliant. The number that predicts whether AI will still be in use in two years is the share of former sceptics who use it on their own work without being watched. Do not count the enthusiasts. Ask the sceptics, and ask them by name.
The third is putting the enthusiast in front of the sceptic. After keynotes, the person who waits until the room has emptied to ask me a question is rarely the enthusiast; it is the head of department who wants to know whether she is allowed to say no. Pairing her with the most excited person in the building teaches her that the organisation has taken a side. Pair her instead with someone who was sceptical last year and can say what changed.
What good looks like
A school that has brought its sceptics with it is recognisable by who is doing the talking. Here is a hypothetical, drawn from the pattern rather than any one place. The maths department was the most reluctant in the building in September; by February the second in maths is presenting the results of a marking trial she was asked to run precisely because she did not believe in it, and her figures include the week it went wrong. The list of what the school will not use AI for was drafted by the English teacher who wrote to the head objecting to the whole programme, and it is on the website with her wording. The first reported AI error came from a member of staff who had been silent in the first INSET, and it was thanked in the staff briefing. Nobody has been converted. Four people were answered, one was employed, and the tool is in use in departments that were never on the pilot.
Contrast the school where, on paper, everything went well. Adoption is reported at 70%. The enthusiasts run a Friday club. The sceptics are described as "coming round" and have not been asked a direct question since September. The list of what the school will not do does not exist, because writing it felt negative. That school will report the same 70% next year, from the same people, and will call it maturity.
Where to start this fortnight
Start by asking three people the Sceptic's Question, and start with the three you least want to ask. Not in a survey: in a corridor, one at a time, with no agenda to persuade. Write down the four words their answers point to, purpose, permission, proof or protection, and take the tally to the next leadership meeting as the year's first piece of adoption data.
Then supply the thing most often named. If it is purpose, the leadership team owes the staff one sentence about what AI is for here this year. If it is permission, the head owes them a page, read aloud. If it is proof, the next pilot needs a sceptic inside it and a date. If it is protection, someone senior owes the staffroom the story of their own mistake.
Finally, find your principled objector and offer her the pen. If she declines, ask again in a term. If she accepts, you have just recruited the most credible author your AI policy will ever have.
If your leadership team is trying to bring a divided staff with it this year, this is the work I do with schools and trusts through AI strategy sessions and implementation support.
Sources and further reading
- EdChoice and Morning Consult, Surveying Teachers on Artificial Intelligence, from the Spring 2026 Teacher Survey (1,031 US teachers, fieldwork 1 to 9 April 2026), published 7 May 2026. Full report: Teachers and K-12 Education: Spring 2026.
- Gallup and the Walton Family Foundation, Most Teachers Receive No Formal Guidance on AI Use, from Teaching for Tomorrow: Closing the Expectations Gap (2,069 US K-12 teachers, fieldwork 9 February to 2 March 2026), published 26 May 2026.
- Teacher Tapp, AI for planning, leadership gaps plus TAs, 21 April 2026.
- METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 10 July 2025.
- Anne Lutz Fernandez, Resisting AI Mania in Schools, Rethinking Schools, volume 40 number 2, winter 2025-26.
- David Timis and Shaista Khilji, The 5 faces of human readiness for AI adoption, World Economic Forum, 1 June 2026.
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 more than thirty countries. About Dan.
