Should students use AI in class?

If the tool elevates learning, use it. Otherwise don't.

This sounds self-evident until you look at how these decisions actually get made. In many schools, decisions about AI tools are made in the IT department, in the leadership team, or by a single enthusiast on the staff — and the question being asked is "should we roll out AI?" That's the wrong question. The right one is "for this class to learn this, does AI elevate learning or not?" The answer may well be no for one class and yes for another, or no for one topic and yes for another — and it needs to be the teacher's decision, because the teacher is the only one with all the relevant facts on the table.

In April 2025 the OECD published Unlocking High-Quality Teaching. It is not an AI report — it is a report on what teaching actually is, grounded in 150 classrooms across 40 countries. The key finding is worth quoting: "In an era defined by rapid innovation and constant change, it is tempting to focus on the latest trends or technologies that promise transformative change. However, refining existing teaching practices by closely examining the current realities of classrooms can be a powerful — and even potentially safer — approach to addressing stagnating student achievement." No other factor inside a school, the report concludes, has a greater impact on student outcomes than the quality of teaching. No technology is going to change that.

Challenging thinking, or replacing it

When students use AI, they can do it in two fundamentally different ways that look superficially alike but have entirely different effects on learning. The first is to let AI replace thinking — "write my essay on World War II", "solve these maths problems", "give me three arguments for and against nuclear power". The result is a finished product, delivered without the student doing the cognitive work that learning consists of. The second way is to let AI challenge thinking — "here's my answer, explain why you think the reasoning is off", "I've tried this problem but I'm stuck, which step is most likely wrong?", "I've drafted an op-ed, which counter-arguments are strongest and how would I address them?" The result is that the student works harder, not less, but with a sparring partner that raises the ceiling on the work.

The research is currently mixed, and it's worth being honest about that. In one corner stands a substantial 2025 meta-analysis (Wang & Fan, Computers & Education, 69 studies) showing that ChatGPT in controlled experiments tends to improve students' performance, motivation and higher-order thinking in the short term. In the other corner stands a set of more recent studies pointing to long-term effects in the opposite direction. An EEG study from MIT Media Lab 2025 (Kosmyna et al.) measured brain connectivity while participants wrote essays with or without LLM help — the LLM group had up to 55 percent lower brain connectivity. A parallel study from Microsoft Research (Lee et al., CHI 2025) of 319 knowledge workers showed that high confidence in AI goes hand in hand with less critical thinking. A Swiss study (Gerlich 2025) found a correlation of -0.75 between AI use and critical thinking among younger users.

What does that add up to? Probably that AI makes it easier to perform on today's task while making it harder to build the cognitive architecture required for future tasks. That is a reasonable assessment, not a proven truth — but it should suffice as a precautionary principle in a school context. A student who got good marks in year 9 but can no longer think their way through a hard problem on their own isn't a success story.

Teacher knowledge isn't a bonus — it's a prerequisite

The most ignored fact in the AI-in-schools debate is that many teachers who are meant to let students use AI in their lessons have themselves never had solid AI training. That's not a complaint about teachers — it's an observation of a systemic gap. In many cases, education authorities have purchased AI tools and made them available to the staff without allocating time for the upskilling those tools require to be used responsibly. The outcome is predictable: students get to use tools whose output the teacher cannot fully evaluate.

In 2024 UNESCO published its AI Competency Framework for Teachers. It defines 15 competencies across five dimensions — among them a human-centred mindset, AI ethics, foundational AI understanding, AI pedagogy, and AI to support one's own learning. The framework doesn't say every teacher must be an AI expert. It says there is a baseline that is a prerequisite for responsibly letting students use AI — and that baseline is not reached by clicking through a pre-recorded 30-minute introduction.

Teachers don't need to know the tech better than students

A common excuse for not bringing AI into lessons is "the students know this stuff better than I do." That's often true in a superficial sense — a 14-year-old can navigate an interface faster than many adults — but it's wrong in the sense that actually matters. The student doesn't always know what an LLM is. The student rarely knows how to craft a prompt that elicits deeper answers. The student can almost never judge whether the answer is true. And above all: the student doesn't have the teacher's understanding of what the lesson is meant to teach.

That last point is the teacher's expertise, and it's irreplaceable. So the more constructive framing of the same situation is this: teachers don't need to know the tech better than students, but teachers need to know the pedagogy better — and then you can learn the tech together. This works, provided the teacher holds on to the role of setting learning goals and judging results, and uses the students' tech familiarity as a resource in the classroom rather than a threat.

What the 1:1 laptop era taught us

When schools rolled out "one laptop per student" (1:1) in the 2010s, researchers tracked what happened. The research record is clear and internationally consistent. OECD's PISA report Students, Computers and Learning (2015) found what became known as the "digital paradox": students who used computers frequently at school performed worse on PISA, even after controlling for socioeconomic background. The US economics study Carter, Greenberg & Walker (2017) ran a randomised controlled trial at the US Military Academy and found that classes where laptops were permitted had 0.18 standard deviations lower scores than classes where they weren't. A meta-analysis of 1:1 programmes in Review of Educational Research (Zheng et al. 2016) concluded that positive effects existed — but only where teachers integrated the devices pedagogically, not simply because the technology was present.

The most cited Nordic longitudinal study — Grönlund et al., Örebro University, 2011–2014 (the Unos Uno project, covering roughly 20 schools in 11 municipalities) — reached a similar conclusion with striking directness: the schools that improved their results with 1:1 had more teacher-led time, not less. Where the teacher stepped back and let the device "take over," results fell. Where the teacher reinforced their own pedagogical work with the device as an aid, and stayed closer to students' work, results improved. The study also showed that solitary work increased in 1:1 classrooms, that social media distracted a substantial share of students, and that weaker students were hit hardest — precisely the students schools exist to help.

Grönlund's summary line is worth remembering: "A school IT project isn't a tech project, it's a change project." Technology doesn't elevate learning on its own. It amplifies the pedagogical choices the teacher has already made. If those choices are good, they get better; if no choices have been made at all — if students are simply turned loose in a digital environment — you get what the 1:1 research consistently documented: more solitary work, more distraction, worse learning, especially for those already struggling.

A classroom management tool isn't a luxury — it's a prerequisite

Whether or not your school uses AI, it is by now reasonable to recommend that schools acquire a classroom management tool that governs what students can do on their devices during class. This isn't about distrusting students; it's about the teacher's job to create a learning environment where students succeed — and they don't, if they're one click away from TikTok during the lesson. The 1:1 research was explicit on this point: the problem wasn't the students, the problem was that teachers lacked the tools to manage the digital environment in the classroom, which meant responsibility got abdicated to the student.

When AI becomes part of the lesson environment, the need grows. A management tool that can open and close access to specific AI services for specific lessons — so that a student has access to one approved AI tool for a task but not to the whole internet during that time — is a baseline requirement for responsible in-class AI use. Combined with aggregated class visibility (if the vendor offers it), teachers suddenly have the instruments the 1:1 researchers were calling for back in 2014.

Pedagogy

1. What learning should this lesson produce, and does AI elevate that learning or replace the work students need to do themselves? If the second half of the answer is "replace," choose a different tool.
2. Does AI challenge the student's thinking, or replace it? "Explain why my answer is wrong" challenges; "write my essay" replaces.
3. Would the same learning outcome be reached equally well or better without AI? If yes, and AI adds nothing beyond convenience, don't use AI.
4. Does AI use fit the subject and topic? (I in WISE — Inspect what the subject requires.)

Prerequisites

5. Do I as the teacher have enough AI literacy to evaluate students' AI responses and catch bias and hallucinations? If not — wait on using AI for this task until you do, or plan the lesson so you evaluate AI responses together as a class.
6. Have students practised source criticism of AI answers? If not — do a short joint exercise on the large screen before the actual AI lesson.
7. Is the tool compliant with your jurisdiction's data protection rules and approved by your school authority? In the EU this means GDPR compliance with a signed data processing agreement; outside the EU check the equivalent rules (e.g. FERPA in the US). If in doubt — ask your data protection officer or IT coordinator before the lesson, not after.

Oversight

8. Do I have visibility into what students are asking the AI during the lesson — at least at the aggregated class level? If not, the lesson depends on you moving around the classroom, looking over students' shoulders and listening. Plan the time for that.
9. Does the school's management tool or clear rules prevent students from leaving the assigned task? If the school lacks a management tool — use explicit AI time in short blocks and combine with analogue segments.
10. Can I interrupt and hold a short whole-class discussion if I notice the class is stuck on the same point? Plan for the possible interruption in the lesson design from the start.

Follow-up

11. How will I assess whether students actually learned more or less than they would have without AI? (E in WISE — Evaluate the outcome.) Plan a concrete follow-up — a short quiz, an oral presentation, a written reflection — where the student, without AI, has to demonstrate what they've learned.
12. What signs of cognitive outsourcing should I be looking for? E.g.: answers that are too complete and too abstract for the student's level, absence of the typical mistakes the student usually makes, loss of the student's own voice in text.