High school is where three promises meet: homework should help students practise, exams should provide credible evidence of what they can do, and career guidance should help them choose a path without narrowing their future too early. Generative AI challenges all three, but not in the same way. The answer is not to ban every tool or replace every assessment. It is to redesign the connections between practice, proof and planning so that students learn with AI when it helps and can still think, explain and decide without it.
That means a reset of routines, not a retreat from academic standards. Homework can become a place to practise with transparent boundaries; exams can sample unaided knowledge and responsible tool use; and career guidance can use AI to widen exploration while keeping human judgement, context and student agency at the centre.
Why high school needs a different conversation
Teenagers increasingly encounter AI while preparing assignments, revising for tests and imagining future study or work. A single school rule cannot govern every task well. An AI-generated paragraph submitted as independent assessed writing raises an authorship problem. A tutor that gives a hint after a student attempts a difficult algebra problem raises a different question. A chatbot that helps a student compare apprenticeship and university routes may be useful, but its recommendations still need checking.
The OECD’s PISA 2025 results, published on 8 September 2026, underline why schools should avoid simplistic conclusions. The report finds that the relationship between AI use and science performance varies by purpose: students using AI for specific tasks such as drafting or summarising showed a different pattern from students using it more generally to help them learn. These are associations in a large international assessment, not proof that AI caused a particular result. The OECD’s separate Digital Education Outlook 2026 makes the instructional distinction more directly: GenAI can support learning when guided by teaching principles, while outsourcing cognitive work can improve the product without producing learning.
For school leaders, the practical question is not whether students touch AI. It is whether each school task produces the evidence its purpose requires.
Homework: protect practice without turning home into a surveillance zone
Homework is usually practice, preparation or consolidation. It can help students retrieve knowledge, apply a method and notice what they do not yet understand. If AI supplies every answer, the completed page may look stronger while concealing the exact gap the teacher needs to see. If schools prohibit all AI use without teaching students how to question it, they also miss opportunities for guided practice, feedback and access support.
A better homework design names the learning goal and the permitted role of assistance. For example, a mathematics task might ask students to try a problem unaided first, request one hint if stuck, then solve a similar problem without help and explain the method. A history assignment might allow students to ask AI for counterarguments only after they have chosen and supported their own claim. A language task might permit vocabulary suggestions while asking the student to identify which suggestions changed the intended meaning.
These routines preserve a visible student contribution. They also make the boundary easier to explain: AI may support the route to learning, but the student still has to perform the learning.
Schools should keep documentation proportionate. A routine practice task may need only a brief note about AI assistance and one independent check. A major assessed project may require more detailed records, according to local rules. Requiring screenshots of every prompt for ordinary homework can create compliance work without improving learning, and may collect unnecessary personal information. A short explanation of what the tool did, what the student checked and what the student can now do is often more educationally useful.
Homework should also not assume every student has equal access to approved devices, paid tools, quiet space or reliable connectivity. If a task depends on a particular AI service, provide an equivalent school-based route. Otherwise, a policy intended to modernise homework can reward access at home rather than learning.
Exams and assessment: separate the conditions, then decide what each result means
The phrase “AI-proof assessment” suggests a contest that schools cannot reliably win. Students can use tools outside the exam room; detection systems can be uncertain; and a polished final product alone may not reveal who made the important decisions. A more defensible design combines tasks under different conditions and states what each one is meant to measure.
An unaided checkpoint can show whether students recall essential knowledge, choose a method and explain core reasoning without assistance. An AI-permitted task can assess a different capability: whether students can question generated material, verify claims, improve a draft or use feedback responsibly. A short oral explanation, annotated decision or transfer question can help connect the two. These elements need not be added to every assignment. The evidence burden should rise with the stakes and with the difficulty of attributing work to an individual student.
For example, in a science investigation, a learner might use AI to suggest possible variables but must distinguish suggestions from observed data, cite the sources used and explain the final design. In an English essay, students might compare an AI-generated interpretation with textual evidence, then defend their own reading. In a timed examination, supervised unaided work may remain appropriate because the qualification is intended to show individual performance under shared conditions.
Rules for formal qualifications are jurisdiction-specific. In England, Ofqual’s guidance for the 2027 assessment season, published on 7 September 2026, tells schools and colleges to follow awarding-organisation instructions for non-exam assessment and says students must not submit AI-produced work as their own; permitted use must be acknowledged. That is a concrete English example, not a universal rule for every country or exam board. Schools elsewhere should use the current requirements of their own authorities.
The design test is validity: what can a teacher or examiner legitimately infer from the result? A high mark on AI-assisted work cannot automatically stand for independent mastery. Nor should a student’s ability to write without digital tools be mistaken for the only capability worth assessing. Schools need both a clear baseline and opportunities to assess competent, critical use where that is part of the learning goal.
Career guidance: use AI to widen the map, not choose the destination
Career guidance is often reduced to a one-off meeting or a list of occupations. For high school students, good guidance is an ongoing process of discovering interests, learning about pathways, testing assumptions and making revisable decisions. AI could help students generate questions, compare course requirements or explore unfamiliar roles. It may make information easier to access outside school hours, but availability is not the same as accurate or equitable guidance.
The OECD’s 2024 policy paper on digital technologies in youth career guidance identifies potential gains in reach and efficiency while warning that benefits cannot be assumed. Its subject is digital guidance broadly, not proof that generative AI can replace counsellors. The OECD’s 2026 Digital Education Outlook notes that well-tuned GenAI may support study and career guidance, but this should be treated as a design possibility, not an outcome guaranteed by a chatbot.
A school can use AI as a first-pass exploration partner, with safeguards. Students might ask it to list several routes into a field, then verify entry requirements against current university, apprenticeship or government sources. They might compare the everyday tasks of two occupations, then discuss what fits their interests and circumstances with a counsellor, teacher or trusted adult. Guidance should help students ask better questions, not make algorithmic recommendations that appear to decide who they are or what they are capable of.
Bias and missing context matter. A model can repeat stereotypes about gender, class, disability or prestige; omit local routes; or describe an occupation using outdated labour-market information. Students should be taught to ask: What source supports this claim? Which routes are missing? What would change this recommendation? Who can help me test it in the real world? Human advisers remain important for interpreting family circumstances, wellbeing, access needs and the uncertainty of a young person’s developing interests.
A practical three-part reset for school leaders
Leaders can begin with a small review of one subject, one year group and one recurring task in each area.
Map the learning claim. For every selected homework, exam or guidance activity, write down what the school wants the student to learn or demonstrate. If the purpose is unclear, changing the AI rule will not fix the task.
Set the assistance boundary. State whether AI is prohibited, permitted for a limited role, or deliberately included. Explain how students should acknowledge use and where the relevant formal assessment rules apply.
Choose evidence that fits the claim. For homework, collect a small unaided attempt or explanation. For assessment, combine an appropriate product with a check of reasoning or transfer. For career guidance, require students to verify information and reflect on choices rather than accept a generated recommendation.
Check fairness and workload. Ask whether students have equivalent access, whether accommodations are respected, whether staff can apply the rule consistently and whether the evidence process takes more time than it returns.
Review what changed. Compare the quality of student explanations, the usefulness of feedback, the clarity of assessment decisions and the confidence of students discussing pathways. Treat early observations as local evidence, not proof that the policy will work everywhere.
This sequence also prevents three common errors. First, it avoids using AI detection as a substitute for task design. Second, it avoids treating every AI interaction as misconduct. Third, it avoids assuming that a student who can operate a tool has learned the subject or made a sound career choice.
Schools can build from existing work on assessing learning when students use AI and checking whether school AI leads to more learning. Those questions connect to a broader goal: keeping students responsible for their thinking while teaching them to use new tools with judgement.
The reset is about better evidence and wider agency
High school should not preserve every homework convention simply because it is familiar, and it should not abandon exams simply because AI can produce convincing work. It should distinguish practice from assessment, assisted performance from independent mastery, and career exploration from career prescription. Each task can then state the role of AI, the work that remains the student’s and the evidence a teacher needs.
A thoughtful reset gives students structured opportunities to use AI, clear reasons for when they must work independently, and repeated chances to explain their decisions. It gives teachers usable evidence rather than more paperwork, and gives counsellors tools that support rather than replace human guidance. Most of all, it prepares young people not merely to obtain answers, but to learn, judge and choose when answers are easy to generate.
Planning a school leadership or education conference on AI, assessment and student agency? Explore Learning Coach’s international keynotes and request a tailored proposal.
References
OECD. PISA 2025 Results (Volume I): Future-Ready Students. Published 8 September 2026. International assessment report; AI-use and performance findings are observational associations, not causal estimates.
OECD. OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. Published 19 January 2026. Synthesis of emerging evidence and policy considerations.
OECD. Digital technologies in career guidance for youth: Opportunities and challenges. Published 4 December 2024. Policy paper on digital career guidance, not a test of GenAI replacing human advisers.
Ofqual. What people working in schools and colleges need to know before exams and assessments. Published 7 September 2026; applies to England.
UNESCO. Guidance for generative AI in education and research. Published 7 September 2023; updated 16 January 2026. Human-centred guidance, not a country-specific assessment rule.
Sources checked 29 October 2026. This article offers an international school-design framework; formal assessment and data-protection rules vary by jurisdiction.
