AI in education: 86% of pupils use it, 65% of teachers know nothing about it

The figures that put an end to the debate

The debate over "should AI be allowed in schools" is out of date. The 2025-2026 figures make it even slightly absurd. According to several converging studies (HEPI, Gallup-Walton Family Foundation, Chegg, OECD), 86% of students across 16 countries use AI in their studies. The annual growth is dizzying: from 66% in 2024 to 92% in 2025 for regular use. The proportion of students using ChatGPT for their assessments has risen from 53% last year to 88% this year.

Among teachers, adoption is following suit. 83% of K-12 teachers (the American equivalent of primary through secondary school) use generative AI, mainly for planning, feedback to pupils and creating content. 53% use AI chatbots (such as ChatGPT or Google Gemini) with their pupils at least once a week. And 60% of educators say they use AI for routine daily tasks.

But here is the paradox: only 13% of schools have a formal policy on AI. The rest are navigating by sight, somewhere between resigned tolerance, unenforced bans and unchecked adoption. This article gives an honest assessment of how schools are changing in practice, following on from our article on the UCLA scandal over AI-generated textbooks.

How pupils really use AI

The reality of use is a long way from the cliché of the "lazy pupil who gets ChatGPT to copy out their essay". The HEPI study (Higher Education Policy Institute, a British higher education research body) has documented precisely what students are doing in 2025-2026:

• 53% use AI to find information quickly (the equivalent of an improved Google search)
• 37% use it to get started on their homework (generating an outline, ideas, a structure)
• 33% use it to get quick explanations of difficult concepts
• 30% use it to rephrase or improve their own writing
• 28% consider it acceptable for ChatGPT to write their work entirely

That last figure is the most problematic. More than one student in four considers it normal to submit work entirely generated by AI. It is on this boundary that the pedagogical debate is being fought. No one really objects to a student using AI to understand a concept. Everyone recognises that using AI to produce an essay submitted as one's own poses a problem.

The problem for teachers: AI content detection tools have become virtually useless. GPTZero, Turnitin AI Detector and others have false positive rates (genuine human texts flagged as AI) of 10 to 30%, and false negative rates (AI texts not detected) exceeding 50% on recent models such as GPT-5 or Claude Opus 4.7. Several universities have officially abandoned these tools.

The impact on teachers: the measured time saving

On the teachers' side, the main measured benefit is time. Teachers who use AI at least once a week save 5.9 hours a week on average, according to a 2025 Gallup-Walton study. Over a school year, that represents around 6 weeks of "recovered" time.

What are those 6 hours gained used for? The main uses among teachers:

• 44% use AI for research and gathering content
• 38% use it to create lesson plans
• 38% use it to summarise information
• 37% use it to generate classroom materials
• A third use it to create assessments and mark papers
• A growing number use it to draft emails to parents and handle administrative communications

As we explored in our article on AI solutions for SMEs, this administrative time saving is one of the most measurable benefits of AI in 2026, whatever the sector. For teachers, it is particularly valuable: they can spend more time with their pupils instead of filling in paperwork.

The problem: 60% of teachers and pupils have received no AI training. 41% of teachers say they do not feel prepared to integrate AI into their curriculum. 65% of teachers still do not use generative AI, and 9% do not even know it exists. The divide between AI-native teachers and the rest is widening fast.

The areas where AI really works well

Not all educational uses are equal. Here are the ones that produce genuinely measurable results.

Personalised learning. Khan Academy, one of the largest free educational platforms in the world, rolled out its Khanmigo assistant in 2023. By 2026, it serves 4.2 million daily unique users. The principle: every pupil has access to a personalised AI tutor that adapts to their pace, level and weak points. The preliminary results are impressive for motivated pupils, but the effect is weak for disengaged pupils (AI does not invent missing motivation).

Language support. For learning a foreign language, AI is probably the major pedagogical revolution of the decade. Unlimited conversation, instant correction, immersion at zero cost. Duolingo (the leading language-learning app, based in Pittsburgh, valued at more than $17 billion on the stock market) completely redesigned its product around AI in 2024-2025, with personalised dialogues generated on the fly.

Accessibility. 90% of educators believe AI can improve accessibility and personalised learning. For dyslexic, visually impaired, hearing-impaired and autistic pupils, AI provides tailor-made tools that did not exist 5 years ago. This is probably the area where the impact is clearest and least controversial.

Science and mathematics. Current reasoning models (Claude Opus 4.7, GPT-5.5, Gemini 3) now solve problems at the level of the International Mathematical Olympiad (the IMO, the most prestigious competition for secondary-school pupils in mathematics). As we documented in our article on scientific discoveries by AI, mathematical reasoning capabilities have exploded. For a sixth-former or a science student, having access to a tutor of that calibre for free (via the free versions of ChatGPT, Claude, Gemini) is probably the biggest educational revolution in recent history.

The areas where AI causes problems

Not everything is rosy. Several problematic uses are emerging and deserve attention.

Creative writing and argumentation. When a student uses ChatGPT to write their philosophy essay, they miss the point of the exercise. The purpose of an essay is not to produce a text, it is to structure one's own thinking. If AI does the work, the student does not learn to think. Several literature and philosophy teachers are radically rethinking their assessments to get around this problem: oral exams, in-class writing without devices, defending one's arguments. This is probably the most profound transformation of pedagogy currently under way.

Learning through effort. Part of learning comes from cognitive struggle: trying, failing, starting again. If AI instantly gives the answer, that process disappears. Several studies are cause for concern: students who use AI intensively remember less well and develop their critical thinking less. This is what researchers call "cognitive offloading" (in French: externalisation cognitive, the act of delegating one's reasoning to an external tool).

Factual accuracy. AI models still "hallucinate", meaning they sometimes invent false information presented as true. For a student who does not check, this is a dangerous trap. Several academic papers have already been published with citations invented by ChatGPT. As we documented in our article on school textbooks, this problem also affects teachers who trust AI blindly to prepare their lessons.

Equity of access. 80% of secondary school pupils receive AI training, but only 8% of pupils from nursery to Year 3 (Pre-K to 3rd grade in the USA). This inequality of access and training could widen existing inequalities. Pupils from affluent families with educated parents make better use of AI, at a younger age, with more discernment. The others fall behind.

How smart teachers are adapting

Beyond the philosophical debate, several teachers have already devised pedagogies that integrate AI creatively. Three approaches that work.

Approach 1: "Work with the AI, show me how." Instead of banning ChatGPT for homework, some teachers ask pupils to document their dialogue with the AI. What questions they asked, what answers they received, how they verified and used them. Assessment focuses on the quality of the interaction, not the final result. This is probably the most innovative emerging pedagogy in 2026.

Approach 2: spoken and written work in class. Many universities are returning to in-person and oral assessments. If the student has to defend their argument live, AI cannot help them in real time. This approach restores the assessment of real skills, but it is costly in terms of organisation time and logistics.

Approach 3: long, personalised projects. Rather than a standardised essay, some courses require unique projects (analysis of a specific work, local case study, original creation). AI can help with the basics, but cannot replace the personal engagement that a bespoke project demands.

The impact on teaching professions

Contrary to fears, teachers are not going to be replaced by AI in the short term. Projections from the US Bureau of Labor Statistics forecast growth in university jobs of 24% between 2025 and 2030. K-12 teachers remain at stable or slightly growing headcounts.

What is changing is the nature of the job. The skills that are gaining value:

• Learning design (designing learning experiences suited to a world with AI)
• Emotional and relational support (what AI does not do well)
• Motivation and inspiration (getting pupils to enjoy a subject)
• Classroom management and group dynamics
• Qualitative and formative assessment

The skills that are losing value:

• Frontal transmission of knowledge (AI does it better)
• Marking standardised work (AI does it faster)
• Routine administration (AI does it for free)
• Planning repetitive lessons (AI generates them in minutes)

This transition weighs on the least adaptable and least trained teachers, while rewarding those who accept the tool. As we explored in our article on the disruption of graphic designers, it is exactly the same pattern across all intellectual professions in 2026.

The huge market taking shape

AI in education has become a massive market. According to several converging studies:

• The global AI in education market will reach $10.4 billion in 2026
• 31% of that market is in North America
• K-12 schools account for 53% of AI deployments in education (higher education 39%)
• Annual growth exceeds 35% a year

The main players:

• Khan Academy with Khanmigo (free, non-profit model)
• Duolingo for languages (valued at $17 billion)
• OpenAI with ChatGPT Edu (a version dedicated to educational institutions)
• Anthropic with Claude for universities
• Google with Google Workspace for Education and Gemini built in
• Microsoft with Microsoft 365 Education and Copilot
• Coursera, edX, Udemy, which are integrating AI heavily into their online courses

In France, several startups are positioning themselves: Educlever, Mathia, Studapart. But the market remains dominated by American players, as we explored in our article on France falling behind.

The specifically French debate

In France, the debate on AI in schools is more polarised than in the United States. The Ministry of National Education published a framework for AI use in April 2024 that remains cautious. Several academies (Paris, Versailles, Aix-Marseille) have launched supervised experiments. But no clear national policy is emerging.

The result: practices are largely informal. Teachers who want to use AI do so on their own initiative, without official training. Pupils use ChatGPT massively (French figures follow international ones, around 80-85% of sixth-formers and students), often without their teachers knowing.

Several specific debates are stirring French schools:

• The French baccalauréat oral exam. How do you assess an oral skill when the pupil can practise with an AI? The Grand Oral of the bac (an exam introduced in 2021 in which the pupil presents a personal project) is particularly affected.
• The brevet and standardised assessments. The DNB (Diplôme National du Brevet, the end-of-collège exam in France) includes written papers that are becoming hard to tell apart from AI-assisted work.
• Philosophy in the final year. The most emblematic subject in the French school system is also the one where AI is most effective at producing convincing essays. Philosophy teachers are in the middle of a rethink.

What to take away for 2026

Three honest observations by way of conclusion.

Observation 1: adoption has already happened. The debate over "should ChatGPT be banned" is as relevant as "should calculators be banned" was in the 1980s. The tool exists, it is used on a massive scale, and continuing to fight it head-on is probably pointless. The real question is how to integrate it intelligently.

Observation 2: training is the weak link. 60% of educators and 60% of pupils have received no training. The danger is not AI itself, it is unsupervised use of AI. Investing heavily in teacher training and pupils' digital literacy is probably the number one education policy priority for the next five years.

Observation 3: good pupils will benefit enormously, others not as much. The Harvard study on tutored learning shows a mixed result: it is motivated pupils who get the most out of it. Disengaged or struggling pupils are not automatically helped by AI. The risk is that this technology widens gaps instead of closing them. It is probably the biggest pedagogical challenge of the coming years.

The French school system is behind, but not disastrously so. It still has a few years to structure a coherent response. For parents and teachers reading this article, the practical advice: rather than banning it, teach your children and your pupils to use AI intelligently. Check sources, do not delegate everything, understand the limits, keep your critical thinking. That is what will genuinely prepare them for the world of 2030, not an outright ban.

To go further, you can read our article on AI solutions for SMEs, which shares similar issues around professional adoption, and our analysis of scientific discoveries made by AI, which shows how far teaching could go with these tools.