AI in education in 2026: use cases, real examples and what comes next

Picture a group of high school students in Turkey sitting a maths exam after weeks of practising with a ChatGPT-style tutor. The ones who had used the chatbot freely had done far better in practice. On the exam, without it, they did worse than classmates who never touched it. That one result captures the whole debate about AI in education better than any slogan.

Here is the short answer. AI in education means using software that can generate text, explain ideas, mark work and adapt to a learner, for tutoring, lesson planning, feedback and admin. The evidence so far says it helps when it is designed to make students think, and hurts when it simply does the thinking for them. Most students already use it; schools are still working out the rules.

The short answer, with the numbers

Students using AI is no longer a fringe habit. A Pew Research Center survey of 1,458 US teens, fielded in autumn 2025 and published in February 2026, found that 54% have used chatbots to help with schoolwork, and about one in ten use them for all or most of it. In UK universities the figure is close to universal: the HEPI and Kortext Student Generative AI Survey 2026 reports that 95% of undergraduates use AI in some form and 94% use it to support assessed work.

Teachers have moved too, though more slowly. A Gallup and Walton Family Foundation poll of 2,232 US public school teachers, reported by The 74, found that most used AI in the 2024-25 school year and that weekly users saved almost six hours a week. The same poll found that 68% had received no AI training and only 19% worked at a school with an AI policy. A smaller 2026 survey of Michigan educators by Michigan Virtual put professional and personal use above 80%, yet teachers' trust in AI fell compared with 2025.

So the picture in late 2026 is lopsided: high use, low training, thin policy, and wobbly trust. That gap is where most of the real problems sit.

How AI in education works

Schools have used software that adapts to learners for decades. Maths programs that serve an easier question after a wrong answer were around long before ChatGPT. What changed in late 2022 was the arrival of large language models: systems trained on huge amounts of text that predict, word by word, a plausible response to whatever you type. They can explain photosynthesis five different ways, write a quiz, or comment on an essay draft in seconds.

In practice, AI turns up in schools in four main roles:

• Tutor. A chatbot that explains, asks questions back and gives hints. We cover this in more depth in our explainer on what an AI tutor is and what the evidence says.
• Teacher's assistant. Drafting lesson plans, worksheets, differentiated reading levels, emails to parents and rubrics.
• Feedback and assessment. Comments on drafts, practice questions and, more cautiously, help with marking.
• Homework helper. Photo-to-answer apps such as Gauth (see our Gauth AI review) and the subject solvers in our guide to science AI solvers.

The technology underneath these roles is often the same. What differs is the design around it. A general chatbot will happily hand over a finished answer. A tutoring tool can be told to withhold the answer, ask a guiding question and only confirm when the student gets there. That design choice, not the model, turns out to matter most.

Real examples of AI in classrooms

Nigeria: an after-school English tutor

In June and July 2024, Edo Boys High School in Benin City ran a six-week after-school programme in which students worked with a generative AI tutor, supported by teachers. The World Bank team behind the randomised evaluation reported gains of about 0.3 standard deviations, which they described as "nearly two years of typical learning in just six weeks". They also said the programme outperformed 80% of the evaluated education interventions in developing countries in the database they compared it against, and that girls, who started behind, gained more. Two cautions: the study measured mostly English, and attendance was disrupted by flooding and strikes.

Harvard: an AI tutor versus active learning

In a randomised trial published in Scientific Reports in June 2025, students in an introductory Harvard physics course did one lesson with a purpose-built AI tutor at home and another in a well-run active-learning class. The median learning gain in the AI lesson was more than double the in-class gain, and students spent a median of 49 minutes on it. The tutor was carefully scripted to follow good teaching practice, and the students were Harvard undergraduates, so this is a best-case setting rather than a typical one.

Turkey: the warning from maths class

The study from the opening scene, by researchers at the Wharton School and published in PNAS, involved nearly 1,000 high school maths students. According to Knowledge at Wharton, students with an unrestricted ChatGPT-style tool did 48% better during practice but 17% worse on the later exam without AI. A version with guardrails, which gave hints instead of answers, lifted practice scores by 127% and left exam scores about level with the control group. In other words, the guardrails removed the harm but did not, in this study, add lasting gains.

Whole systems: Estonia and New York City

Countries and cities are now making very different bets. Estonia's AI Leap programme, launched in 2025, has given tens of thousands of upper-secondary students access to AI tools, including an Estonian version of ChatGPT designed to guide students through problems rather than hand over answers, and trained thousands of teachers. New York City went the other way for younger pupils: in September 2026 Mayor Zohran Mamdani announced a one-year moratorium on student-facing generative AI from kindergarten to eighth grade, with companion chatbots banned in all grades, AI literacy classes for high schoolers, and teachers still allowed to use AI for planning and admin.

Benefits and limitations

An analogy helps here. AI in education is a bit like a forklift in a gym. If your goal is to move boxes, a forklift is brilliant. If your goal is to get stronger, it defeats the point. Schoolwork is mostly the gym kind of task: the essay or the solved equation matters less than the change it makes in the student's head. The OECD's Digital Education Outlook 2026 makes the same point in plainer terms, warning that better outputs with general-purpose tools can "vanish or reverse" when AI is taken away, and naming the risk of "metacognitive laziness".

Putting the studies above side by side shows the pattern clearly.

Study	Tool design	Result while using AI	Result without AI
Turkey, high school maths (PNAS)	Open chatbot, no guardrails	48% better in practice	17% worse on exam
Turkey, high school maths (PNAS)	Tutor with hints, not answers	127% better in practice	About the same as control
Harvard, intro physics (Scientific Reports)	Scripted tutor built on teaching research	Not applicable	Median learning gain more than double active learning
Nigeria, after-school English (World Bank)	Chatbot tutor with teacher support	Not applicable	About 0.3 SD gain in six weeks

Our reading of this table: the best results come from tools built to teach, used with a teacher nearby. The worst come from open chatbots used as answer machines.

What AI does well

• Time for teachers. Gallup's weekly users reported saving almost six hours a week, mostly on preparation and paperwork.
• Patient, on-demand help. A tutor that never sighs at the fifth "can you explain that again?" is useful, especially for students who cannot afford private tutoring.
• Adapting material. Rewriting a text at a lower reading level or into another language takes seconds.
• Better support from novices. The OECD notes that less experienced tutors improved their support, and student outcomes, with education-specific AI tools.

Where it falls short

• Thinking skills. In the Gallup poll, 57% of teachers expected weekly AI use to reduce students' independent thinking.
• Cheating and trust. Pew found 59% of teens say AI cheating happens at their school at least somewhat often. Detection is no fix: as our explainer on how AI detectors work sets out, accuracy is contested and false accusations happen.
• Errors. Language models can state wrong facts with confidence, which is a real problem for a learner who cannot yet spot the mistake.
• Privacy and policy gaps. Few schools have clear rules, and student data flows to companies whose terms change.

What it means for you

The practical advice depends on who you are, but it follows one principle: use AI to make the thinking easier to start, not to skip it.

• Students. Ask for hints, explanations and practice questions rather than finished answers. Try the problem first, then check. Our list of the best free AI tools for students and our guide to math AI tools note which tools have step-by-step or tutor modes. Always check your school's rules before using AI on assessed work.
• Teachers. Planning and admin are the lowest-risk, highest-reward uses. For student-facing use, prefer tools you can configure to withhold answers, and consider assessments that show the process, such as drafts, oral explanations or in-class work. US K-12 teachers can also get ChatGPT for Teachers, which OpenAI launched in November 2025 as a free, verified workspace and says does not use shared content to train its models by default.
• Parents. Ask your child to explain how they got an answer. If they can, AI probably helped them learn. If they cannot, it probably did the work.
• School leaders. Write a policy, train staff, and say clearly which uses count as cheating. Students in the HEPI survey said AI skills matter for their future, yet "fewer than half feel supported by their teaching staff".

What happens next

What we know: use will keep growing, and the research is converging on design. What we are guessing: whether schools will split into two camps, with some leaning into supervised AI tutoring and others restricting it for younger children.

That split is already visible in the US. Newsweek's October 2026 roundup lists New York City's moratorium, a proposed New York state bill (A9190) that would bar most classroom AI before ninth grade, and Los Angeles Unified's limits on student generative AI on school devices. At the same time, Maryland, Oklahoma, Idaho and Virginia passed laws in 2026 that set guardrails rather than bans, such as required district policies, AI coordinators and parental opt-outs.

Our best guess is that the debate will shift from "AI or no AI" to "which AI, for which age, doing what". Tutors that ask questions rather than answer them, teacher-facing tools for planning, and AI literacy lessons look set to become normal. Open chatbots for young children look set to face more limits. And the question of whether AI will replace teachers looks less likely than the question of how teachers use it to do the parts of the job only they can do.

FAQ

How does AI support education?

AI supports education by giving students on-demand explanations, hints and practice questions, and by saving teachers time on lesson planning, worksheets and admin. In a Gallup poll, US teachers who used AI weekly reported saving almost six hours a week. The benefits are strongest when tools are designed to guide students rather than give answers.

How is AI used in education?

AI is used in four main ways: as a tutor that explains and asks questions, as a teacher's assistant for planning and differentiation, for feedback on drafts and practice, and as a homework helper. Pew found 54% of US teens have used chatbots for schoolwork, most often for research, maths problems and editing their writing.

Why should AI be used in schools?

The case for AI in schools rests on access and time. A well-designed AI tutor can give every student patient, one-to-one help, which trials in Nigeria and at Harvard linked to strong learning gains. Teachers can hand off routine preparation. Students also need AI literacy, since most already use these tools outside class.

Why is AI bad in schools?

AI can be bad in schools when students use it to skip the thinking. In a study of nearly 1,000 Turkish maths students, those with an unrestricted chatbot scored 17% worse on an exam without it. Other concerns include cheating, confident factual errors, unreliable AI detectors, student privacy and a lack of clear school policies.

How can AI help with education?

AI can help most when you use it like a tutor rather than a shortcut. Try a problem first, then ask for a hint or an explanation of your mistake. Ask it to quiz you before a test or to simplify a hard reading. Teachers can use it to draft materials, then check and adapt them before use.