ThinkYoung’s research covers 5,000 teenagers and 600 teachers, while OpenAI’s country accounts document reported time savings, student frustrations and unanswered questions about learning outcomes
OpenAI’s Responsible AI in Education report highlights the gap between students’ AI homework use and the guidance they receive.
OpenAI is calling on governments and schools to strengthen teacher training and classroom guidance after ThinkYoung research found that 79% of surveyed teenagers had used AI for homework or assignments, but only 47% had been shown how.
Its Responsible AI in Education publication draws on education partnerships in Estonia, Slovakia, Greece and Italy. It sets out priorities for teacher preparation, critical AI literacy and evaluation, while acknowledging that adoption and reported time savings do not establish whether students are learning more.
Through Education for Countries, OpenAI is working with governments on access to AI tools, jointly developed literacy programs and evaluation adapted to local languages, curricula and institutions.
Writing on LinkedIn, George Osborne, Head of OpenAI for Countries, said: “Our ambition is for AI to strengthen what young people can understand, create and do for themselves, while giving teachers more time for thought, creativity and individual attention.”
ThinkYoung’s research provides a detailed account of the classroom practices those programs are seeking to address. Alongside widespread use and confidence, it records uncertain permission, limited instruction in recognizing unreliable answers and students’ willingness to copy AI-generated work.
Students ask for clearer rules and help checking AI answers
The student survey reached 5,000 secondary school students aged 13 to 19 across 10 European countries: Austria, Estonia, France, Germany, Greece, Italy, Poland, Spain, Sweden and the UK. Responses were collected between April 14 and May 15, 2026.
A separate survey covered 600 teachers across 10 countries, including Kenya and the United Arab Emirates. The combined survey coverage therefore spans 12 countries. Eight focus groups added the experiences of 52 students, parents and teachers.
Among surveyed teenagers, 95% reported using AI tools generally. For schoolwork, students who used AI most commonly sought clarification, looked up concepts and generated answers to questions. Just over a quarter, 26%, reported using it to do an assignment for them.
Permission was frequently unclear. Among students who had used AI for schoolwork, 43% said they had teacher permission the last time they used it, 36% reported using it without permission and 21% were unsure.
Students without clear guidance were twice as likely to report asking AI to do their work for them as those receiving guidance, at 36% compared with 18%. These are associations in survey responses rather than evidence that guidance caused the difference.
Instruction in checking AI content was less common than general discussion of the technology. While 75% said teachers discussed AI’s opportunities and challenges, only 35% reported being shown how to identify bias or incorrect information.
Students’ descriptions of their own habits also reveal tensions. Some 88% said they added their own thoughts to AI outputs and 74% said they checked accuracy. Yet 49% agreed that they would simply copy what AI told them when completing homework.
For that separate copying question, ThinkYoung’s detailed analysis found no significant influence from teacher permission or guidance. Students with a positive view of AI’s educational impact were more likely to report copying its answers.
One 15-year-old student in Belgium told researchers: “I think it’s important that you ask AI specifically how to do things, so AI does not just give you the answer.”
Students were asking for more help in making those decisions. A majority wanted teachers to explain when AI could be used for homework, at 55%, and help them judge whether its outputs were reliable, at 56%. A smaller share, 41%, wanted teachers to use AI more in class.
Younger teenagers expressed particular uncertainty. Among 13- to 15-year-olds, 65% wanted clearer homework guidance. Their confidence in judging AI accuracy was also lower: 45% felt confident, compared with 61% of 18- to 19-year-olds.
Half of respondents identified AI’s risks as something they wanted to learn about at school. Its possibilities and societal effects were also priorities, each selected by 40%.
Teachers report workload benefits alongside concerns about student thinking
AI use was already common among the teachers surveyed. Some 80% used it for work or workload support, and 52% used it while teaching a class or lesson.
Much of that activity operated outside institutional accounts. Some 62% reported accessing AI through a personal account unaffiliated with their school or education institution.
Teachers most commonly used the tools to create classroom content and plan lessons. Saving time was identified as a main benefit by 65%, while 47% highlighted support for creative or engaging teaching.
Only 14% identified freeing up time for face-to-face interaction as a main benefit. The responses therefore give stronger support to reported workload relief than to claims that AI is already creating more individual attention for students.
Privacy and security concerns were the most frequently reported barrier to wider use, selected by 48%. A lack of clear school guidance followed at 34%.
A 29-year-old teacher in the UK told the focus groups: “If AI is going to be incorporated much more into education, there should be trainings and workshops organised for teachers.”
Teachers also wanted products more closely matched to their work. Their leading requests included better tools for tasks such as lesson planning and student feedback, more accurate outputs, and tools tailored to teachers and their curricula.
Views on student benefits remained mixed. Some 85% agreed that AI helped students with additional learning needs, but these responses represent teachers’ perceptions rather than independently measured improvements.
The potential loss of critical thinking was teachers’ leading concern about AI’s future educational impact, selected by 58%. Other concerns included reduced creativity, overreliance and academic integrity.
Country programs produce mixed early results
OpenAI’s publication organizes its partnership lessons around five priorities: educational goals, educator preparation, local expertise, critical AI literacy and credible evidence. It presents the experiences as context-specific lessons rather than a single approach for every education system.
Estonia’s AI Leap program illustrates the uneven early results. In a spring survey, 794 of 828 teachers reported professional AI use, while 723 estimated average weekly savings of 1.8 hours.
The ITI tutor, built on ChatGPT Edu, recorded average satisfaction of 3.07 out of five among 182 students. Feedback cited excessive questioning, slow exchanges and demand for clearer explanations. These responses do not establish its effect on learning.
Greece’s framework requires trained teacher oversight and school AI coordinators, and prohibits fully automated assessment decisions and profiling. Slovakia’s approach brings its education ministry together with universities and technical partner Aricoma.
In Malta’s AIforAll program, participants must complete three learning modules before receiving a free premium AI subscription. Its first phase serves the wider population.
OpenAI also calls for independent evaluations that can publish unfavorable findings without supplier veto. The University of Tartu and Stanford’s SCALE are designing and validating its Learning Outcomes Measurement Suite.
The company says its OpenAI Learning Lab is bringing education researchers together to investigate when and how AI supports education systems. Its stated research priorities extend beyond immediate use and satisfaction to knowledge, independent thinking and retention.
ThinkYoung’s recommendations give schools several practical starting points. It calls for age-appropriate rules covering homework, revision, classroom activities and assessments, with clear explanations of when AI is encouraged, restricted or inappropriate.
It also recommends embedding accuracy checks, bias recognition and misinformation awareness into curricula, supported by subject-specific teacher training. Parents and carers should receive practical resources explaining school expectations, privacy and responsible use.
For assessments where AI is permitted, ThinkYoung recommends requiring students to document their initial aims, prompts, responses to AI-generated material and final accuracy checks. Marking should recognize reasoning, application of concepts and the quality of students’ critique alongside the finished work.
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.
