The honest question in education bothering educators and school leaders alike.
I know what it feels like to stand in front of a room and teach something you’re not sure you understand. Every educator does. You learn the lesson the night before. You stay one chapter ahead. You get comfortable with the discomfort, and you teach anyway, because that’s the job.
But this is different. This is the first time in my professional life that I can remember a technology arriving in classrooms faster than teachers could learn it, faster than curricula could absorb it, faster than institutions could form a position on it. And the gap between what students know and what their teachers are equipped to teach is not, for once, a gap we can quietly close before anyone notices.
Students have already noticed.
| 65% | Of higher education students believe they know more about AI than their instructors. A further 68% believe their teacher’s AI competency is a matter of luck rather than institutional preparation. (Cengage AI in Education Report 2025; Vodafone Stiftung / Frontiers in Education, 2025.) |
Read that again. Not a small fraction. Not a vocal minority. Two-thirds of students think they are ahead of the people teaching them. And when nearly seven in ten believe that whether their teacher knows anything useful about AI is essentially a coin flip, that is not a perception problem. That is a structural failure being experienced in real time, in real classrooms, by real students who are making decisions about AI every day without meaningful adult guidance.
So here is the question I can’t stop sitting with: what does it actually mean to teach something you don’t fully understand yourself? And is that necessarily as catastrophic as it sounds?
What the data is telling us
The numbers on teacher AI preparation are stark. In the United States, seven in ten teachers had received no AI training at all as of spring 2024. By fall 2025, that figure had improved, roughly half of teachers had attended at least one professional development session. But one session is not readiness. Of those who had received training, 41 percent rated it “poor” or “mediocre.” Only 18 percent called it “good” or “excellent.”
The US is not an outlier. It is, if anything, better resourced than most. The OECD’s Teaching and Learning International Survey, covering 32 countries in 2024, found that teacher AI use ranged from 14 percent in France to 52 percent in Albania; a gap that reflects not motivation but the presence or absence of national training infrastructure. Globally, fewer than 10 percent of schools have established any formal guidelines for generative AI. Sixty-eight percent of urban teachers worldwide have had no AI training at all.
And yet 86 percent of students globally are already using AI for their studies. They are ahead of us, and they know it, and most of them are navigating it without any structured guidance about how it works, what its limits are, or what it means to use it honestly.
| >80% | Of students in the US were NOT explicitly taught how to use AI for schoolwork by their teachers, despite more than half of teachers now using AI themselves. (RAND Corporation, 2025; nationally representative survey.) |
That figure, more than eight in ten students not being taught about something they are already using daily, is the real headline. Not whether teachers are using AI. Not whether schools have a policy. Students are navigating one of the most significant technological shifts of their lifetimes largely alone.
And the adults responsible for preparing them are, for the most part, figuring it out at the same time, though not in the ways students need. As education professionals, we are channelling considerable energy into how AI can optimise our own work: lesson planning, curriculum mapping, email correspondence, report writing. That focus is understandable. But it risks becoming a form of negligence if it crowds out the harder question of how students are using it right now, and what they understand about what they’re doing.
I have watched this shift in real time. In 2024, AI detection platforms could identify AI-generated text with reasonable accuracy; a useful check on student work. By 2026, that has changed significantly. I recently tested several platforms by submitting texts written entirely by AI. The plagiarism results came back near zero. I was not surprised; I had expected it. AI has come a long way since those first staff room conversations in 2021, when a sixth-grade essay on climate change, complete with academic citations, gave itself away almost immediately. Today, that is no longer the case.
And even teacher engagement with AI is uneven. What schools “think” all their teachers are doing with AI is often quite different from what is actually happening, shaped by school type, district funding, leadership priorities, and access to professional learning. A system cannot close a gap it refuses to accurately map.
| “I was asking for a district policy for student use of AI last spring and was brushed off. Teachers shouldn’t be left out in the wind on this issue.” |
— High school teacher, open-ended survey response. EdWeek Research Center, 2026.
This teacher is not unusual. The EdWeek Research Center’s surveys consistently find that lack of knowledge and support is one of the top reasons teachers aren’t using or teaching AI. Not resistance. Not laziness. Not fear of change. A system that handed them a technology and offered no infrastructure to support its use.
This is not a teacher problem
Every article I have read that frames this as a question of teacher readiness gets it slightly wrong. Teachers are not the ones who decided to deploy AI into schools without a preparation infrastructure. Teachers are not the ones who left professional development underfunded and optional. Teachers are not the ones who allowed a technology adopted by hundreds of millions of students to exist, in most schools and most countries, in a policy vacuum.
Researchers at Teachers College, Columbia University put it plainly: there is an understanding of what AI literacy is, but almost no understanding of what it looks like in practice across different disciplines and age groups. There is no scaffolding. That is an institutional failure masquerading as an individual one, and we should be honest about the difference.
The equity dimension deepens this. In the US, low-poverty school districts are nearly twice as likely to have provided AI training to teachers as high-poverty ones. In the UK, students in disadvantaged schools are significantly less likely than their peers to receive any structured teaching on how AI works. Globally, the countries most likely to have national AI teacher training programmes are the countries that were already the most educationally resourced. The AI preparation gap is not separate from the existing educational equity gap. It is the same gap, accelerating.
What if not knowing is the point?
Research on how learning actually develops does not suggest that students need to observe expert mastery. It suggests they need to observe expert thinking. There is a meaningful difference. Faculty who openly display intellectual humility, who acknowledge gaps in their knowledge, invite alternative viewpoints, and model how to navigate uncertainty, produce students with stronger critical thinking and greater resilience. The teacher who says “I don’t fully understand this yet either, and here is how I am thinking about it” is not admitting defeat. They are teaching something that no AI can demonstrate: what it looks like for a thoughtful human to sit with uncertainty and reason through it carefully.
If we do not model this ourselves, we risk producing students who turn to technology to perform confidence rather than develop it. That is not what education is for, and it is not what AI was designed to enable.
A 2025 study of an experimental undergraduate AI unit found that framing the teacher’s own uncertainty as a pedagogical resource, what researchers called “educator vulnerability as method,” produced measurably better critical AI engagement in students than approaches built on delivered expertise. The students who watched their teachers figure it out alongside them became better at interrogating AI outputs, not worse.
This matters because what we are trying to produce is not students who know how to use AI tools. We are trying to produce students who know how to think about them, who can interrogate outputs, recognise bias, understand what AI cannot do, and make informed decisions about when and how to use it. Those are not technical skills. They are intellectual ones. And they are best modelled, not delivered.
| 68% | Of students globally believe their teacher’s AI competency depends on chance rather than systematic preparation. But research shows that teachers who model intellectual humility and uncertainty positively influence student outcomes — even when they don’t have answers. (Vodafone Stiftung / Frontiers in Education, 2025; DeVries, Orona & Arum, 2025.) |
The questions that still need answers
Reframing teacher uncertainty as a pedagogical resource is not a licence for systems to continue under-preparing teachers. It is an argument for a different kind of preparation, one that builds confidence not through the pretence of expertise, but through genuine engagement with AI: its possibilities, its failures, its ethical dimensions, and its limits.
There are questions that remain genuinely open, that no reframe resolves. What happens when students navigating AI without guidance make decisions with lasting consequences about academic integrity, about critical thinking habits, about which sources they trust? What happens to students in under-resourced schools, in lower-income countries, in systems where the AI preparation gap is not closing but widening? Is “modelling uncertainty” a meaningful pedagogy when the uncertainty includes not knowing whether the technology is reinforcing bias, narrowing perspectives, or quietly disadvantaging certain learners?
I recently spoke with a family who were using AI to audit their children’s extracurricular profiles against college admissions criteria; adding and removing activities entirely on the basis of AI recommendations. That deserves its own article. But it points to something this piece is really about: we cannot underestimate how deeply, how variably, and how uncritically students and families are already relying on AI to make consequential decisions. The question of whether students use AI has been settled. The question of whether they understand what they are using has not.
I don’t have clean answers to any of this. I am not sure anyone does yet. But I think these are the right questions; more honest, and more useful, than the one we tend to ask in public, which is usually some version of: should students be allowed to use AI at all?
That question has already been answered. They are using it. The question now is who is helping them understand what they are using, and why, and what it means, and what to do when it gets it wrong.
And right now, in most classrooms, in most countries, the honest answer is: nobody.