Perhaps PISA 2025 is telling us what really matters.

By Gavin Keller

A warning in the data

The latest PISA results should make everyone involved in education stop and think. More than 760,000 15-year-olds across 91 countries and economies participated in PISA 2025, representing around 33 million young people worldwide. Across OECD countries, average performance reached its lowest recorded levels in science, mathematics and reading. Reading has fallen by 28 points since 2015 and mathematics by 22 points. Science has proved more resilient, particularly between 2022 and 2025, but its longer-term trajectory has also weakened.

My first reaction was probably the same as that of many educators: What has gone wrong with education? But the longer I looked at the graphs, the more another question bothered me: What if we have been asking the wrong question?

The great measurement experiment

The first PISA assessment took place in 2000, at roughly the same historical moment that education systems around the world were embracing standards, accountability, targets, testing, benchmarking and measurable outcomes. The belief was understandable: better data would identify problems, targets would focus attention, accountability would produce action, and improved performance would follow.

Twenty-five years later, education has never possessed more data. We can compare countries, provinces, districts, schools, subjects, teachers, cohorts and demographic groups – sometimes even individual children to two decimal places. Yet PISA 2025 reports the lowest OECD-average performance observed across its three core domains. This does not prove that PISA, testing or assessment caused the decline. There are far too many variables for that conclusion. But it does give us permission to examine the theory of change beneath our obsession with measurement.

When the measure becomes the mission

Gert Biesta warned in Good Education in an Age of Measurement that measurable outcomes can begin to define what we believe education is for. Dylan Wiliam, whose work has shown the power of formative assessment, has likewise highlighted the limitations of using standardised test results to make simplistic judgements about school quality. Campbell’s Law and Goodhart’s Law offer a similar warning: once an indicator carries serious consequences, people begin adapting their behaviour to the indicator itself.

Consider a report showing MATHEMATICS: 82.37%. It looks wonderfully precise. But what does it tell us about the learner’s misconceptions, transferable understanding or next learning step? What meaningful educational difference exists between 82.37% and 82.14%? The number answers one question particularly well: Who won? Learning asks a different question: What does this learner understand, and what should happen next?

South Africa tried something different

I should declare my bias. I was involved in South Africa’s experiment with Outcomes-Based Education and Curriculum 2005. Before becoming a principal in 1993, I had already implemented Project-Based Learning in two schools, and I introduced it at Sun Valley Primary. I was later asked to represent the teacher union NUE, subsequently NAPTOSA, on the design team working on South Africa’s Outcomes-Based Education model.

One idea I valued deeply was its assessment philosophy. The intended learning was clearly established and evidence of attainment could be represented on a four-point scale. 

A 3 or 4 indicated that the learning had been attained. 

A 1 or 2 did not mean, ‘This child has failed.’ 

It meant, ‘This has not been achieved yet.’ That word yet changes assessment because the result becomes information for the teacher: What should we change? Another explanation? A different scaffold? More background knowledge, practice or time? Then teach again and gather evidence again.

Assessment is not formative because of the instrument. It becomes formative because of what the teacher does next. A four-point scale can become just as dysfunctional as a percentage if its purpose is simply to rank children. The magic was never in the numbers. The magic was in the response.

Performance is not learning

A learner can perform without deeply learning. A science student may reproduce content brilliantly in an examination yet struggle to use it in a novel situation. A child can pass mathematics while remaining terrified of mathematics. A learner can achieve highly in reading while never choosing to read.

This matters because PISA itself increasingly looks beyond recall towards scientific literacy, problem solving, evaluating evidence and making connections. Its 2025 reading findings point to declines in capabilities that matter enormously in the AI age: evaluating information, connecting multiple sources and critically interpreting what is read. Deep learning reveals itself when knowledge can be recognised, connected, applied and transferred beyond the original lesson.

Don’t blame PISA

It would be easy to make PISA the villain. That would be a mistake. PISA is a sophisticated attempt to obtain credible, comparable evidence across profoundly different education systems. It examines far more than national averages, including proficiency distributions, equity, socio-economic factors and variation within systems.

The problem begins when governments, media and systems demand that PISA means more than it can legitimately tell us. A complex population-level assessment becomes a league table, and the complexity of learning is reduced to who came first.

What should happen next?

The OECD’s response to the 2025 results is strikingly familiar: focus on fewer areas in greater depth, invest in teachers, engage parents, provide targeted support where it is needed most, and use technology and AI purposefully rather than allowing them to replace attention, effort and understanding.

Perhaps PISA has not failed at all. Perhaps it has done precisely what it was designed to do: give us evidence. The next twenty-five years may require us to become as sophisticated about how humans learn as we have become about measuring performance.

Not less evidence, but better use of evidence. Not no assessment, but assessment that drives the next learning decision. Not abandoning knowledge, but knowledge that produces meaning and transfer. Not removing accountability, but accountability for learning rather than simply for numbers.

Perhaps the greatest lesson from PISA 2025 is not contained in the international rankings at all. We repeatedly discover what matters in learning – and then retreat to what is easiest to measure. Twenty-five years after the first PISA assessment, perhaps it is time to resist that retreat.