Artificial intelligence is no longer a future issue for schools. Teachers are already using it to produce resources, plan lessons and complete administrative work, while pupils are using it to explain topics, generate ideas, complete homework and sometimes avoid doing the thinking themselves.
A 2026 National Education Union survey of more than 9,000 teachers found that 76% were using AI for their day-to-day work. However, 49% said their school had no AI policy for either staff or pupils, while 66% reported having no policy specifically covering pupil use. AI adoption appears to be moving faster than school policies, staff training and the evidence about its effect on learning.
The question is therefore no longer whether AI will enter education. It already has. The challenge is deciding where it improves teaching, where it threatens learning and where human judgement must remain firmly in control.
This article focuses mainly on schools in England, although many of the principles apply more widely.
1. Move beyond the choice between banning and embracing AI
Schools are sometimes presented with two extreme options: ban generative AI completely or encourage everyone to use it.
Neither position is particularly realistic.
A complete ban is difficult to enforce because pupils and staff can access AI tools outside the school network and through an expanding range of websites, apps and devices. Unrestricted adoption is equally problematic because the tools can produce inaccurate, biased, generic or inappropriate material.
Schools need managed use. This means identifying acceptable purposes, setting boundaries and reviewing whether AI is genuinely improving education.
The starting question should not be, “How can we use more AI?” It should be, “Which problem are we trying to solve?”
2. Begin with teaching and learning, not the technology
A school should not introduce an AI tool simply because it looks impressive.
Technology should serve a clear educational purpose. It might help a teacher generate retrieval questions, adapt the reading level of an existing text or create examples for pupils to critique. It might allow pupils to compare two explanations or identify weaknesses in an AI-generated argument.
However, using AI is not automatically innovative or effective. A quickly generated worksheet can still be a poor worksheet. A personalised explanation can still contain misconceptions. An automated activity can still waste lesson time.
The quality of the teaching decision matters more than the novelty of the tool.
3. Use AI to produce starting points, not finished teaching
AI can generate lesson ideas, quizzes, model paragraphs, explanations and questions in seconds. This speed is useful, but it can also create the illusion that the work is complete.
Teachers still need to check:
- whether the content is accurate;
- whether the level is appropriate;
- whether the examples match the curriculum;
- whether the questions assess the intended knowledge;
- whether the language is accessible;
- whether the resource contains stereotypes or assumptions;
- and whether the activity will actually help pupils learn.
The Department for Education describes generative AI as a possible starting point for creating resources such as lessons, quizzes, revision activities and communications. The important phrase is “starting point”. Professional review remains essential.
4. Target repetitive workload rather than professional thinking
One of the strongest arguments for AI is its potential to reduce workload.
An Education Endowment Foundation trial involving 259 teachers in 68 secondary schools found that teachers using ChatGPT with an implementation guide reduced their weekly planning time for Year 7 and Year 8 science by 31%. Teachers used it mainly to create questions, generate activity ideas and adapt existing materials.
This is promising, but it does not mean schools should automate every part of planning.
AI is likely to be most useful for repetitive first-draft tasks: producing question variations, reformatting existing material, simplifying instructions or generating possible examples. Teachers should remain responsible for sequencing learning, anticipating misconceptions and deciding what their classes need.
The goal should be to remove avoidable administration, not remove the intellectual work of teaching.
5. Do not confuse faster feedback with better feedback
Generative AI can produce detailed feedback almost instantly. That does not necessarily make the feedback accurate, manageable or useful.
Automated comments can be too long, too vague or based on a misunderstanding of the pupil’s reasoning. They may correct surface features while missing the central misconception. Pupils may also ignore large amounts of impersonal feedback.
AI may help a teacher identify common errors, organise notes or draft possible comments. However, teachers need to decide which feedback will move learning forward.
In formal qualification assessments, JCQ guidance states that teachers must not use AI tools as the sole marker of pupils’ work.
AI may support professional judgement. It should not replace it.
6. Create one clear policy for staff and pupils
A policy should not be a vague statement telling everyone to use AI “responsibly”.
Staff and pupils need practical answers:
- Which tools are approved?
- Which age groups may use them?
- When is AI use permitted?
- When must permission be requested?
- What information must never be entered?
- How should AI assistance be acknowledged?
- What counts as academic misconduct?
- Who should staff contact when they are uncertain?
- What happens when a tool changes its terms or privacy arrangements?
The policy should cover staff use, pupil use, assessment, safeguarding, data protection and procurement. It should also be written in language that pupils and parents can understand.
Having separate, contradictory guidance from different departments will create confusion and make enforcement unfair.
7. Define acceptable use for individual tasks
“AI is allowed” and “AI is banned” are rarely sufficient instructions.
Acceptable use depends on the purpose of the task. A teacher might allow pupils to use AI to generate three possible essay questions but not to write the essay. Pupils might ask AI to explain a concept after attempting it independently, but not use it to complete a graded assessment.
Schools could use a simple task-based system:
- No AI: The pupil must work independently.
- AI for support: AI may explain vocabulary, suggest ideas or provide practice.
- AI for collaboration: Pupils may use AI but must evaluate, edit and acknowledge its contribution.
- AI expected: The task specifically assesses the pupil’s ability to use AI critically.
This removes ambiguity. Pupils should know the rules before beginning the work, not after they have been accused of breaking them.
8. Teach pupils how generative AI actually works
AI literacy should involve more than learning how to write an effective prompt.
Pupils need to understand that a language model generates likely responses from patterns in data. It does not understand a topic in the same way a person does, and it does not automatically distinguish truth from plausibility.
Pupils should learn that AI can:
- invent facts and references;
- reproduce bias from its data;
- provide different answers to similar questions;
- express false information confidently;
- oversimplify disputed issues;
- and conceal uncertainty behind fluent language.
AI literacy is becoming part of literacy itself. Pupils must learn to question the source, evidence, purpose and reliability of machine-generated content.
9. Make verification a normal classroom habit
AI-generated information should never be treated as automatically trustworthy.
Teachers can model verification by asking pupils to check claims against textbooks, official data, reliable websites and original sources. Pupils could highlight every factual claim in an AI answer and find evidence supporting or contradicting it.
This turns one of AI’s greatest weaknesses into a useful learning opportunity.
The Department for Education advises schools to fact-check AI outputs, while JCQ requires pupils to verify sources supplied by AI when using them in assessed work.
“Check the AI” should become as routine as “show your working” or “cite your source”.
10. Keep personal information out of public AI tools
Staff should never paste identifiable pupil information into an AI system simply because it saves time.
Names, behaviour records, medical information, SEND details, assessment data, safeguarding concerns and family circumstances are personal data. Free tools may store prompts, collect additional technical information or use entered material to improve their systems.
The Department for Education advises staff to consult their data protection officer or IT lead, use approved tools and understand how data is processed. Its July 2026 guidance warns that entering pupils’ personal information into an unsuitable system could result in a data breach.
A teacher could ask AI to improve the wording of a generic parent email. They should not include the pupil’s name, diagnosis, behaviour history or other identifying details in the prompt.
Convenience does not override data protection.
11. Check for bias rather than assuming neutrality
AI-generated material may reproduce stereotypes or reflect gaps in its training data.
A careers activity might associate certain jobs with particular genders. A history response might prioritise dominant perspectives. An image generator may produce narrow representations of families, scientists or school leaders.
Teachers should examine:
- who is represented;
- who is missing;
- which perspectives are treated as authoritative;
- whether assumptions are being made about culture, gender, disability or social class;
- and whether the output could disadvantage particular pupils.
Bias does not always appear as an obviously offensive statement. It may appear through repeated patterns, omissions or apparently neutral examples.
AI should be subjected to at least the same scrutiny as a textbook, commercial resource or website.
12. Protect productive struggle
Learning often involves effort.
Pupils need to retrieve information, plan answers, select methods, make mistakes and revise their thinking. If AI completes each difficult stage, pupils may produce better-looking work without developing the knowledge behind it.
In June 2026, the EEF announced a major research programme investigating whether generative AI supports learning or encourages pupils to “offload” processes such as recall, reasoning, planning and drafting. It noted that robust evidence about AI’s effect on learners under 16 was still at an early stage, with the first findings from the new programme expected in 2027.
Schools should therefore be cautious about using AI to remove every difficulty.
The best use of AI may be to provide a hint, question or worked example rather than an immediate finished answer.
13. Redesign homework rather than relying on trust alone
Traditional homework can be completed by AI in seconds.
Simply telling pupils not to use it will not always work, particularly when the task asks for a generic essay, explanation or set of factual answers.
Homework could place greater emphasis on:
- preparing for an in-class discussion;
- applying ideas to a local or personal example;
- annotating a source;
- producing a plan rather than a finished answer;
- explaining decisions orally;
- correcting a flawed response;
- completing part of the work in class;
- or comparing an AI answer with a pupil’s own answer.
This does not mean every homework task must become elaborate. It means teachers need to consider whether the task still provides reliable evidence of independent learning.
AI has exposed weaknesses in homework and assessment design that were already present.
14. Collect evidence of the learning process
A polished final product is becoming less reliable as the only evidence of what a pupil knows.
Teachers may need to see more of the process: plans, drafts, notes, classroom writing, source records, version histories and conversations about the work.
Pupils should be able to explain:
- what they were trying to achieve;
- why they made particular choices;
- which sources they used;
- how their work changed;
- where AI contributed;
- and what they accepted or rejected from its response.
JCQ advises centres to use supervised work, portfolios and discussions with pupils when authenticating assessment. Pupils using AI in qualification work may also need to retain their prompts and generated output.
Assessment should increasingly capture thinking, not just submission.
15. Make qualification rules explicit
The rules for ordinary classwork and formal qualification assessment are not always the same.
JCQ requires centres to have a malpractice policy that covers AI use, including when it is permitted, how it should be acknowledged and how misuse will be treated. Work submitted for qualification assessment must be the pupil’s own. Submitting AI-generated material as independent work may constitute malpractice and can lead to serious sanctions.
Ofqual’s updated July 2026 approach to AI in qualifications places emphasis on fairness, validity, security, public confidence and enabling safe innovation.
Schools should not assume that pupils understand these distinctions. The rules should be taught, revisited and communicated to parents before assessed work begins.
16. Do not treat AI detection scores as proof
Schools may be tempted to use AI detection software as a quick answer to suspected misuse.
These tools can form part of an investigation, but they should not be treated as conclusive. Their accuracy can vary according to the model used, the amount of AI-generated content, editing by the pupil, the type of prompt and the pupil’s command of English.
JCQ advises that detection tools should form only one part of a wider judgement. Teachers should consider the pupil’s previous work, classroom performance, drafts, references and ability to discuss the submitted material.
A percentage produced by software is not the same as proof.
Schools need fair procedures that allow concerns to be investigated without automatically accusing pupils.
17. Invest in training and review the policy regularly
An AI policy will achieve little if staff do not understand it.
Training should give teachers opportunities to test approved tools, examine inaccurate outputs, consider subject-specific uses and discuss realistic assessment scenarios. Staff also need guidance on privacy, safeguarding, copyright, bias and the limits of AI detection.
The Department for Education updated its school and college training materials for the 2026–27 academic year in May 2026. These include modules on understanding AI, interacting with generative AI, safe use and educational use cases.
Training should not consist of one enthusiastic demonstration followed by an expectation that every teacher adopts the technology.
Schools should review their approach regularly because tools, pupil behaviour, regulations and evidence are changing rapidly. Responsibility should sit with a named leader or working group, with input from teachers, safeguarding staff, data protection officers, pupils, governors and parents.
Schools need thoughtful adoption, not rapid adoption
AI can help teachers generate ideas, adapt resources and reduce parts of their workload. It may eventually support feedback, tutoring and more personalised learning.
It can also produce misinformation, expose personal data, reinforce bias and allow pupils to submit work they do not understand. Used badly, it can make teaching quicker while making learning weaker.
The central principle should be simple: AI should support human thinking, not replace it.
Schools do not need to predict every future development. They do need clear rules, informed staff, protected pupil data and assessment systems that continue to show what pupils genuinely know and can do.
The most successful schools will not necessarily be those using the most artificial intelligence. They will be those making the most intelligent decisions about when not to use it.