What changed
Based on reporting by The Verge, OECD PISA data collected in 2025 found that 15-year-olds who never used AI generally scored better in science than AI users after socioeconomic adjustment. The study covered more than 760,000 students in 91 countries.
The split is in the use case. AI used to summarise assigned reading or draft assignments tracked with relatively worse results; weekly use for general learning slightly beat non-use, and regular learning use paired with frequent evaluation of AI-generated information beat non-use.
Why This Matters
The useful distinction is not “AI in school: good or bad.” It is whether the tool replaces the work that teaches, or gives someone more material to test, challenge and improve.
That changes what a credible education product should do. A fluent answer box is cheap to build. A workflow that makes users compare claims, spot gaps, check a draft and explain a revision is harder, but it is closer to the pattern associated here with better performance. Andreas Schleicher, the OECD’s director of learning and skills, puts it neatly: learning requires productive cognitive struggle, not simply consuming content.
There is one important uncertainty: these are associations, not proof that AI itself caused the score differences. Still, the direction is a serious warning for anyone treating generated summaries and finished drafts as a learning feature rather than a shortcut with a shiny interface.
Access matters too. The report says AI use was more common among relatively advantaged students, likely reflecting differences in hardware, reliable internet and paid tools. An AI-dependent classroom can therefore make a tool more central just as it makes equal participation less certain.
How the effects could spread
Schools could shift from asking whether AI is allowed to asking students to show how they evaluated it. That would move classroom time toward checking claims, explaining reasoning and revising generated work.
If that shift takes hold, students may use AI more as a research aide and less as a machine for bypassing reading or writing. Education products would then face pressure to support verification and active problem-solving, not merely quick completion.
The chain breaks if AI remains mainly a summariser of assigned texts or drafter of submissions, if teachers lack time and guidance to assess verification work, or if unequal access to devices, connectivity and paid tools persists.
Impact assessment
Students using AI for summaries or drafted assignments are the most exposed now. Those uses were associated with relatively worse performance, suggesting that convenience can displace the very reading, writing and problem-solving practice an assignment is meant to create.
Teachers who routinely require students to assess AI output gain a more practical role than traffic cop. Within weeks, their assignments could make evaluation, correction and revision part of the work itself. That is the instructional condition linked in the report to regular learning-oriented AI users outperforming non-users.
Students with weaker access stand to lose over the next 6 to 12 months if AI-supported coursework expands without equivalent tools and instruction. A laptop, stable connection and paid subscription should not become invisible prerequisites for completing ordinary schoolwork.
Education AI vendors face a mixed outcome over the same horizon. Products built around completed drafts and instant summaries may draw more scrutiny. Products that help users test, compare and revise answers may become more useful, but only if they preserve independent reasoning rather than automate it away.
Scenarios
Most likely
Our outlook (informed speculation): Over the next school year, more schools will treat AI as a guided learning tool rather than an unrestricted shortcut. If schools can give teachers manageable evaluation tasks, assignments will increasingly require students to identify errors, missing sources or revisions in generated material. This is the likeliest path because the reported advantage is concentrated in regular learning-oriented use combined with critical assessment.
That case strengthens if school guidance requires verification or reflection and coursework favours general learning over automated drafting. It weakens if schools respond chiefly with blanket bans, or permit generated work without a checking step.
Upside
If schools pair AI-literacy instruction with broad access to approved tools, regular users could shift toward research, comparison and revision instead of replacing their own work. Students who lack reliable devices or subscriptions would be less likely to be pushed outside the new workflow, and the benefits of guided use would be less concentrated among already advantaged households.
Signs of this path would include shared access to devices, connectivity or approved AI services, plus assignments showing students’ comparisons between AI output and course material. It depends on critical evaluation being built into coursework rather than offered as an optional extra.
Downside
If AI access grows faster than schools’ ability to teach verification and redesign assignments, daily task-completion use could spread. More students could use tools to summarise reading and draft submissions, while teachers become less able to rely on take-home written work as a measure of independent understanding.
This case would be reinforced if daily use for summaries and drafting expands without assessment training. It weakens if students move toward guided research and learning support, and if intentional monthly or weekly use remains more common than daily shortcut use.
What to watch next
Watch whether school policies require students to verify, critique or revise generated material, rather than simply banning AI or allowing unexamined submissions.
Watch the pattern of use. A move toward weekly or monthly general learning use would fit the stronger outcomes in the report; growth in daily summaries and drafting would point the other way.
And watch access. If AI-supported coursework expands, the consequential question is whether schools provide devices, stable connectivity or approved tools to students who do not already have them.
Comments
No comments yet.