PISA 2025: education continues to lose ground

A week ago, from this same platform, we wrote about a unanimous complaint from university teachers: students arriving in their first year of a degree are increasingly unprepared. Their reading and math levels, quite simply, fall short. Along the way, PISA appeared to confirm it, and in that same article we used some of its data to complement the text. In this post we will analyze the OECD report in a little more detail: what does the data say? What lies behind the figures? What is happening with screens and artificial intelligence? What can we learn from the systems that are making progress?

There is a very simple way to sum up PISA 2025. In mathematics, students in OECD countries now score 22 points lower than a decade ago. In reading, 28 points lower. The OECD itself translates these differences into a more understandable scale: a little more than a year of learning lost in mathematics and around a year and a half in reading. In addition, one in five 15-year-olds performs poorly in Science, Mathematics and Reading simultaneously. In 2022 it was 16%.

However, these numbers tell only part of the story.

PISA 2025 arrives at a unique moment. Five years have passed since schools closed because of COVID-19, and three since PISA recorded the biggest collapse in results in its history. At that time, the pandemic accounted for a good part of the explanations. Millions of students had lost months of in-person teaching, education systems had improvised remote learning formulas, and the consequences had been very unequal.

In 2025, we have been left waiting for the rebound. Science, the main focus of this edition, remains roughly stable compared to 2022 in the OECD average. Mathematics loses another nine points and reading, fourteen. In these last two competencies, these are the lowest average results ever recorded by PISA.

The long-term perspective also changes the diagnosis. A good part of the deterioration had begun before 2020. The pandemic accelerated a previous trend and exposed problems with deeper roots: difficulties in guaranteeing foundational learning, teacher shortages in many systems, changes in reading habits, persistent inequalities and a dizzying transformation in the way adolescents access information and relate to knowledge.

Stop here. You have just read approximately 400 words. If you were a 15-year-old, from this point on you would begin to have difficulty understanding the text, keeping your attention, connecting ideas and reconstructing the meaning of what you have read. And we are barely at the introduction.

So, although PISA 2025 speaks of scores, it invites us to ask a much broader question: what is happening with learning in a generation that studies, reads, searches for information and solves problems in a profoundly different way from just ten years ago?

Of the three major competencies measured by PISA, reading is the one that has lost the most ground. And that matters because reading is not just another competency: it is the gateway to almost all the others. You need to read to understand a math problem, follow a scientific explanation, compare two arguments or decide whether a source is reliable.

The data allow us to be even more precise. Between 2018 and 2025, the proportion of readers able to respond accurately and fluently fell by seven percentage points. At the same time, hasty readers (students who move quickly through the text and answer fast, but poorly) rose from around 7% to 11%. Performance also worsens as the test progresses. The OECD interprets both signals as signs of a reduced ability to sustain attention, persevere or maintain confidence when faced with difficult tasks.

And here it is worth clarifying something about those 400 words we have just read. PISA does not say that from word 401 onward an adolescent stops understanding. It does find a greater deterioration with long texts and, above all, in the skills these usually demand: evaluating information, making connections between several sources, and interpreting and reflecting on what has been read. In the OECD average, barely 6% reach the top reading levels, those at which students are expected to be able to understand lengthy texts, handle abstract or counterintuitive concepts and distinguish facts from opinions on the basis of implicit cues.

These are capabilities that are particularly necessary in an environment where information arrives fragmented into messages, videos, headlines, posts and hyperlinks. The OECD observes that the decline in reading for pleasure and the increase in screen time have been accompanied by poorer results, although PISA does not make it possible to establish a causal relationship between the two phenomena.

We have never had so much information at our fingertips, nor so many tools to find it, summarize it and organize it. But the easier it is to access an answer, the more important it is to know what to do with it: to understand it, relate it to what we already know, and question it when necessary. In a world of immediate answers, understanding still takes time.

The distance between socioeconomically advantaged and disadvantaged students has narrowed.

Finally some good news… or is it?

In Science, advantaged students still outperform disadvantaged ones by 85 points in the OECD average. The gap has narrowed since 2022. That is an undeniable fact. But should we celebrate it? No, because the gap has narrowed due to the decline of the former rather than the improvement of the latter. And the same phenomenon appears in Mathematics and Reading.

Understanding this requires precisely one of those operations that PISA tries to measure: relating several pieces of data and resisting the immediate interpretation. A smaller difference between two groups can mean that those at the bottom have risen; it can also mean that those at the top have fallen. The figure is the same. The story, radically different.

Socioeconomic background continues to be one of the main predictors of performance. But its weight varies greatly from one education system to another. PISA identifies a small group (Canada, Estonia, Ireland, Japan, Korea, Macao and the Chinese regions of Beijing, Shanghai, Jiangsu and Zhejiang) that achieves something especially difficult: good results and, at the same time, a relatively smaller influence of family background. In all of them, more than 75% of students reach at least the basic level in science, and socioeconomic status explains less than 10% of the differences in performance.

How do they do it? PISA does not allow us to establish a common recipe or to attribute these results to a specific policy, but it does offer some clues. In several of these systems there are classrooms with fewer interruptions and distractions, greater teacher support and conditions that make it possible to detect and address learning difficulties. Japan, Korea and the participating Chinese regions, for example, combine a high level of discipline in the classroom with very low levels of digital distraction in class. These are very different systems, so it matters less to look for a single formula than to note the result: social disadvantage weighs heavily, but how heavily also depends on what happens inside and around the school.

This is where so-called academic resilience comes in. This is what PISA calls students who belong to the most socioeconomically disadvantaged 25% in their country and yet manage to be among the top-performing 25%. They are the statistical proof of something important: inequality of starting point and inequality of outcomes are not exactly the same thing. Schools cannot erase the circumstances into which a student is born, but some systems manage to make those circumstances explain considerably less about how far a student can go.

Any attempt to explain how adolescents learn today inevitably leads us to technology. PISA 2025 adds an important nuance here: its data support neither the idea that more technology automatically produces more learning nor the idea that screens alone are responsible for its decline.

Moderate use of devices for learning at school is associated with better results. The relationship changes when they are used for leisure during class. Twenty-eight percent of students in OECD countries say that their classmates are distracted by digital devices in most or all science lessons. Those who study in that environment score, on average, 11 points lower in Science, even after accounting for the socioeconomic profile of students and schools. In Spain, where 29% report this distraction, the difference reaches 16 points.

The data thus shift the focus: it matters less how much technology there is in the classroom than how it is used and for what purpose.

Artificial intelligence takes this much further. Forty-six percent of students in the OECD already use AI chatbots at least once a week to help them learn. Students who say they use it to write assignments obtain worse results than those who do not. But among frequent users, a revealing finding emerges: those who have received training in evaluating the quality of AI-generated information tend to perform better than those who use these tools without such guidance.

The difference, though small, introduces an important nuance: artificial intelligence is already part of the way many students learn. The challenge is to teach them to use it without delegating to it what they need to learn to do for themselves.

PISA has begun to explore precisely this terrain with its new Learning in the Digital World assessment. And here it is worth clarifying what it understands by digital competence. The test does not measure whether an adolescent knows how to use a device, but whether they are able to use computational tools to learn something they do not yet know and solve a problem that has no immediate answer.

To do so, students face digital environments in which they have tutorials, examples and feedback. They have to experiment, break a complex problem down into smaller steps, build models or solutions, check whether they work, detect errors and correct them. In short, they must do something much closer to learning than to executing an instruction.

The first results show a significant gap. Sixty-four percent of OECD students reach the level considered proficient in computational problem solving. But only about half manage to combine that ability with basic competencies in science, reading and mathematics. Put another way: there are students who can get by reasonably well with computational tools yet lack some of the fundamental knowledge on which that use should rest.

The distinction is especially relevant amid the expansion of artificial intelligence. Knowing how to use a tool and knowing how to think with it are not the same thing. To build a model you have to understand what it represents; to interpret the data an application returns you have to know what they mean; to detect a wrong answer from an AI you need knowledge against which to check it. Digital competence therefore does not replace Reading, Mathematics or Science. It depends on them and, when used well, allows them to be taken further.

The average is falling, but not everyone is falling. Turkey, for example, improved between 2022 and 2025 in Science, Mathematics and Reading. If we widen the focus to 2015, the proportion of low-performing students there has fallen by 16 percentage points in Mathematics, another 16 in Reading and 25 in Science. Qatar achieves some of its best results ever in 2025 and stands significantly above its scores from a decade ago in all three subjects. Montenegro also improves in all three.

There is no formula that can simply be transferred from one country to another. Nor does PISA allow an improvement to be attributed to a specific policy. But the OECD does point to elements that recur in the systems with the best trajectories: focusing the curriculum so that content is covered in greater depth, supporting teachers, directing resources to those who need them most, involving families and using technology with clear pedagogical objectives.

These are unspectacular conclusions for an era fascinated by disruption. True. But they stand to reason. Because PISA 2025 arrives just as artificial intelligence promises to change the way we search for information, write, calculate and solve problems. Its results are a reminder that taking advantage of that transformation requires strengthening, rather than taking for granted, fundamental capabilities. To ask a machine to summarize a text, you have to be able to judge whether it has understood it. To accept a calculation, you have to understand what it means. To use an answer, you need knowledge against which to check it.

Ten years ago, an adolescent needed to search for many of the answers that today can be obtained in seconds. Ten years from now, they will probably be able to delegate many more intellectual tasks. The great question for education will be deciding which ones it is worth delegating and which ones we must keep teaching precisely because the effort of carrying them out is part of learning.

PISA 2025 thus leaves us with an incontestable truth and a possibility. The reality is that we have been losing ground for a decade. The possibility lies in the systems that manage to move forward while the average falls back, and in a technology that, when well integrated, can expand learning opportunities. The task now is to understand what the former are doing and how to use the latter so that, three years from now, the rebound is no longer something we are still waiting for.