OpenAI’s 10 November 2026 education report highlights a question for secondary schools: are students being taught how to use AI for homework? In ThinkYoung’s survey of 5,000 students aged 13–19 across ten European countries, 79% reported using AI for homework or assignments, while 47% said teachers had shown them how.
ThinkYoung discloses OpenAI’s support for the research. Students answered the survey between 14 April and 15 May 2026; the original report bears an August 2026 copyright date. These are not November responses.
For teachers, students and parents, the practical issue goes beyond access: which parts of an assignment may AI help with, and which thinking must remain the student’s own?
The figures describe reported experiences, not measured learning. Being shown how to use AI for assignments is also narrower than receiving any AI education. Subtracting 47% from 79% would not identify the exact share using AI without guidance: that requires knowing how the same students answered both questions.
The findings should not be extended to all European teenagers or students worldwide. Nor does this observational survey establish that guidance causes better learning or different behaviour.
A finished answer is not proof of learning
A randomised study at one Turkish high school illustrates why assisted work and independent understanding need separate checks. During four mathematics sessions in autumn 2023, nearly 1,000 students used either a basic GPT-4 chat tutor, a tutor equipped with teacher-designed hints, or no AI.
The researchers found that both AI groups did better on assisted practice. On the subsequent unaided exam, however, the basic-chat group did worse than the no-AI group. The hint-based tutor group did not differ statistically from the no-AI group; that is not evidence that it improved unaided learning.
These were short-term findings from one school, one subject and particular tool designs—not a verdict on every AI tutor. They do show why a better homework submission cannot, by itself, establish that a student has learned more.
Four expectations to attach to an assignment
A practical response is to make four things explicit: permission, checking, disclosure and independent understanding. The following approach draws on the research and official guidance; it is a suggested framework, not a tested programme or a promise of better marks.
1. Define the help allowed for this task
Start with the skill being learned. Explain whether students may request hints, explore an explanation, plan an answer, edit their writing or generate a solution. None of these categories should be assumed acceptable for every assignment.
For example, a teacher might permit a hint during practice but require students to write and explain the solution themselves. If constructing an argument is the task, asking AI to supply that argument may replace the very work being assessed. These are illustrative boundaries for a teacher to adapt, not universal rules.
England’s Department for Education guidance invites schools to revisit homework policies and explain when AI use is appropriate, including through engagement with parents. The distinction between access and assignment permission also appears in RohitAI’s guide for Sydney university students, although those university rules are not school policy.
2. Explain what checking an answer involves
An allowed tool can still be wrong. The Joint Council for Qualifications’ assessment guidance requires students to verify AI-suggested sources and independently check content where sources are missing. This is guidance for the qualification assessments it covers, not a worldwide homework rule.
For ordinary homework, make the checking process concrete: open the cited source, check that it supports the claim, and compare the explanation with course materials. For a calculation, work through the steps. Agreement from a second chatbot is not a substitute for checking evidence or reasoning.
3. Make the assistance visible
Ask students to explain what AI contributed and what they changed as a result.
For assessments covered by JCQ’s guidance, effective from September 2026, students must identify the AI source and generation date, retain prompts and outputs, and explain their use. Original work must also be kept where AI modifies it directly. Students still have to demonstrate the assessed skills independently; acknowledgement alone does not earn marks.
Disclosure makes assistance reviewable, not automatically permissible. Follow the actual school and qualification rules.
4. Say what students must demonstrate independently
Identify the reasoning the teacher needs to see. Depending on the task, that might mean explaining one step, attempting a similar problem without AI, or showing how an answer developed. A correct final answer and the ability to explain it are different things.
These are options to adapt, not a requirement to give every pupil an oral exam. Account for accessibility needs and teacher workload. OpenAI’s education report also distinguishes adoption and participant feedback from evidence of educational effects, calling for sustained evaluation.
Give families the same instructions
Put the permitted help, checking expectations, disclosure requirements and independent work on the assignment itself. Share those expectations with parents, so help at home follows the same boundaries. Where instructions are unclear, students should ask the teacher before using AI.
Permission for a task does not settle whether a particular service is suitable for a child. The England guidance also calls for appropriate safeguards and compliance with tools’ age restrictions, and recommends avoiding personal data in generative-AI tools. Schools elsewhere should apply their own relevant requirements.
The immediate decision is not which subscription to buy. It is what the homework is meant to teach—and what the student should be able to do when the chatbot is closed.