AI at school: what students, parents and teachers should know

Is my child allowed to use AI for schoolwork?

It depends on the school, the subject and the task, and there is no global rule. What exists is a patchwork of national guidance, exam board rules and each school's own policy, and they don't always agree.

That patchwork is still young. When UNESCO surveyed more than 450 schools and universities in May 2023, fewer than one in ten had any formal policy on generative AI, the kind of AI that writes text or makes images on request. In September 2023 UNESCO published the first global guidance on it, asking governments to regulate its use in schools and to consider "an age limit of 13" for using it in the classroom, UN News reported.

Since then, many governments have written something down. A few examples show the range:

• England: the Department for Education's guidance, last updated in August 2025, lets schools use AI but tells them to follow the age limits set by each tool, to avoid putting personal data into AI tools, and to review homework policies now that AI is widely available. It also says schools must not let pupils' original work be used to train AI without permission.
• United States: there is no national rule. In July 2025 the federal Department of Education said schools may spend federal grant money on AI tools such as tutoring, while stressing privacy and involving parents. At least 28 states and Washington DC had their own K-12 guidance by July 2025, Stateline reported.
• Australia: education ministers agreed a national framework in October 2023, in force since the start of 2024, built on principles including fairness, transparency and privacy.
• European Union: the AI Act treats AI used to admit students, mark their learning or watch for cheating during tests as "high-risk", which brings extra checks, and it bans AI that tries to read students' emotions in schools, with narrow medical and safety exceptions. Our EU AI Act guide explains when those rules apply.

Rules can also tighten fast. This autumn New York City disabled the student-facing AI features of more than 38 previously approved school programs, saying they did not meet "new safety and oversight standards", Chalkbeat reports. Los Angeles also restricted student use, according to Education Week.

So the only reliable answer for your child is the one in their own school's policy, and in the instructions for each assignment.

What do exam boards say about AI in coursework?

Exams sat in a hall are simple: no AI. Coursework done at home is where the rules matter. The UK's Joint Council for Qualifications (JCQ), which sets rules for GCSE and A-level boards, gives one of the clearest models in its April 2025 guidance.

• Unacknowledged use is malpractice. Copying or paraphrasing AI text, or using AI as a source without saying so, is treated like plagiarism.
• Acknowledged use can be fine, but earns no marks on its own. Students must name the tool and the date, keep a non-editable copy (such as a screenshot) of their questions and the AI's answers, and explain how they used it. Writing just "ChatGPT" in a bibliography is not enough.
• Penalties are serious. They range from a warning to losing marks, disqualification, or being barred from exams for years.
• Teachers must be confident the work is the student's own before accepting it, and a human must always do the marking.

Pressure is growing. In July 2026 the head of England's exams regulator, Ofqual, said written coursework would face "far, far more scrutiny" because of "AI-fuelled subversion of assessment," FE Week reports. Other countries' boards set their own rules, but the JCQ pattern of "declare it, keep a record, and make sure the thinking is yours" is a sensible default anywhere.

Can an AI detector prove a student cheated?

No. AI detectors look for statistical patterns typical of machine-written text, and they make mistakes in both directions: they miss AI text that has been lightly edited, and they flag honest human writing.

The best-known warning came from Stanford researchers in 2023. They ran essays written by non-native English speakers for the TOEFL language test through seven detectors, and the detectors labelled 61% of them as AI-generated, while essays by US eighth-graders were judged almost perfectly, Stanford HAI reports. Simpler, more predictable phrasing looks "machine-like" to these tools, which is a kind of bias against students still learning a language. "Current detectors are clearly unreliable and easily gamed," said James Zou, the study's senior author.

Even OpenAI, the maker of ChatGPT, withdrew its own text detector in 2023, about six months after launch, citing its "low rate of accuracy," AI Business reported. Turnitin, whose checker many schools use, says its false-positive rate is under 1% for documents where it finds 20% or more AI writing, but around 4% for individual highlighted sentences, and it tells teachers to treat highlights as a reason to talk, not as proof, in its own explainer.

The JCQ agrees: detectors can be one check, but never the only basis for a decision. If your child is accused, the strongest defence is evidence of process, such as drafts, notes, version history in Google Docs or Word, and being able to explain the work out loud.

How can students use AI to learn rather than skip learning?

The research points one way: AI helps when it makes you think and hurts when it thinks for you. Our guide on kids, teens and AI chatbots describes a large study where students who got plain answers did worse on the exam later.

The flip side is encouraging. In a randomised study at Harvard, published in Scientific Reports in June 2025, physics students using a custom AI tutor learned more than twice as much as in a well-taught active classroom lesson. But that tutor was built by teachers to guide step by step, with correct solutions written in ahead, rather than left free to answer anything. The authors note the study covered two short topics at one university and measured only short-term results.

Ordinary chatbots can be pushed in that direction. OpenAI added a study mode to ChatGPT in July 2025 that uses hints and questions instead of answers, though the company says it can be inconsistent and can be switched off at any time. Whatever the tool, these habits help:

• Ask for questions, not answers. "Quiz me on this chapter" or "give me a hint, not the solution."
• Try first, then check. Attempt the problem, then ask the AI to find the mistake in your working.
• Ask it to explain differently. A second explanation, a simpler example or an analogy is a good use.
• Check facts against the textbook. Chatbots hallucinate, stating wrong things confidently, including invented quotes and sources.
• Keep the record. Save your questions and the replies, so declaring the use is easy if the assignment allows it.

A good test: after using the AI, could you do the next similar task without it? If not, it did the learning for you.

Which tools are students old enough to use, and what happens to their data?

Consumer chatbots set their own minimum ages. OpenAI's terms require users to be at least 13, with a parent's permission under 18; Anthropic's Claude requires users to be 18. Our kids and chatbots guide covers the age rules and parental controls in more detail.

School accounts are often a different deal. When Google opened its Gemini app to students of all ages through Workspace for Education accounts in June 2025, it said their data would not be reviewed by people or used to train its models, and that school administrators decide who gets access, Chrome Unboxed reported. On personal accounts, chats may be used as training data unless you opt out; our guide to your data and AI chatbots shows where the switches are.

That is a practical reason to use the tool the school provides, when there is one, rather than a personal app.

What can parents ask the school?

You don't need to be an expert to ask good questions. Five cover most of it:

1. What is the AI policy, and where can I read it? Does it differ by subject or year group?
2. How will my child know what is allowed on each assignment? A clear label on each task avoids most disputes.
3. Which AI tools does the school use with pupils, and what happens to their data? Is pupils' work used to train anything?
4. Does the school use AI detectors, and what happens if one flags my child? Ask whether a detector score alone can lead to a penalty, and how a student can show their process.
5. Are pupils taught how AI works and when not to trust it? Knowing that a chatbot is a large language model predicting likely words, not a source of truth, is now a basic school skill.

Expect some honest uncertainty in the answers. A review by Stanford researchers of more than 800 studies of AI in schools found only 20 that gave strong evidence of cause and effect, Education Week reports. Teachers are working this out in real time too.