What PISA 2025 Actually Changed for the Classroom

The OECD published PISA 2025 this morning, and by tonight most coverage will have run the same paragraph: average mathematics scores across OECD countries have fallen 22 points since 2015, reading has fallen 28, and 20% of 15-year-olds are now low performers across all three subjects — up from 16% in 2022. That is real, and there is nothing a single teacher can do about it before Monday.

What is actually useful sits further into the same report. For the first time, PISA asked in detail how students use digital devices and AI chatbots for schoolwork — not screen time as one number, but what for, how much, and with what result. That data is new this year, it is specific, and some of it is not what you’d guess.

The device question got more precise

Every earlier round asked something close to “how much time on a screen.” This one split the question by purpose, and the split is where the finding lives. Limited or moderate use of digital devices for learning activities — up to five hours a day — is associated with higher science scores than either not using them at all or using them heavily. Non-use scores lower too, which the OECD attributes partly to schools that are short on devices or unprepared for digital learning, not to some virtue in going screen-free.

Leisure use at school tells a different story, and the threshold is much lower. Scores are highest among students who use devices for leisure at school for up to an hour a day, and the longer that continues past an hour, the lower the scores. The report’s own framing is closer to a policy answer than a moral one: balancing digital tools with educational and other activities is what determines whether technology helps learning or distracts from it.

This is also the first round to measure classroom distraction directly rather than infer it: 28% of students, on average, said their classmates were distracted by digital devices in most or every science lesson — and the share was consistently lower in schools with a clear phone policy, whether a ban or a well-defined guideline, than in schools without one. We went through this part of the report for our own Digital Wellbeing course, because a school drafting a device policy this term is exactly who this data is for.

The AI data nobody has seen before

AI chatbots only became publicly available in late 2022, so this is the first PISA round that could ask about them properly — and by 2025, a majority of students in most participating countries were already using one for schoolwork.

The headline finding reads like an argument against AI in the classroom: students who don’t use AI chatbots for schoolwork generally score higher than students who do. Look closer and it gets more specific. Weekly, purposeful use of AI to help with learning tracks similarly to not using AI at all — it’s the two extremes, near-daily reliance and occasional, uncommitted use, that score lower than non-users. For narrower tasks like summarising a text or researching a topic, the same shape holds: moderate users outperform both the occasional and the near-daily ones.

The more interesting number sits upstream of all of that. On average, six in ten students report being asked in a lesson to assess the quality of AI-generated information — ranging from under a third of students in some countries to more than four in five in others. And that instruction is not evenly handed out. The OECD’s own conclusion is direct:

“Socio-economically disadvantaged students are less likely to report they are tasked to assess AI-generated information during school lessons… [this warrants] the implementation of targeted interventions, to prevent the emergence of a new form of socio-economic divide in the age of AI.”

That is not a finding about whether to allow AI in a classroom. It’s a finding about who is being taught to question it — and right now, that correlates with the students who were already ahead. Teaching students to evaluate what a chatbot hands them sits squarely inside media literacy, which is most of a week in our Media Literacy course: tracing a claim back to its source, and — since August 2026 — reading what EU law now requires generated content to disclose.

One thing this report does not contain: the full results of “Learning in the Digital World,” the innovative domain PISA built for 2025. The self-regulation component is being analysed separately, in a later report. Today’s volume covers science, reading, mathematics and computational problem solving — the device and AI findings above come from the background questionnaires, not from that assessment.

What to do with this before the next report

• Split “screen time” into two questions, not one. The report’s own finding is that purposeful use for learning behaves differently from leisure use in the same room, and the second has a lower tolerance than most policies assume.
• Add one lesson on assessing AI output, not a new unit. The correlated benefit sits with the instruction, not with the tool — moderate use without it doesn’t show the same advantage.
• Ask who already gets that instruction. If it’s concentrated in your top sets or exam classes, that is the gap PISA is now naming directly.

If your school is drafting what a device policy should actually say, our Digital Wellbeing course works from the European Commission’s own analysis of it, day by day. If the open question is teaching students to question what a chatbot gives them, that’s the Media Literacy course. This is the second time we’ve written about a PISA report — the first was about what Finland’s decline showed about equity — and the pattern holds: the number in the headline is rarely the one worth building a lesson around.

Sources

• OECD, PISA 2025 Results (Volume I), Foreword, 2026, p. 4.
• OECD, PISA 2025 Results (Volume I), Chapter 4 — Student school life and beyond, 2026, pp. 27, 229, 231–233, 237–240, 148.