PISA 2025 and AI at school: when it helps pupils and when it harms them

The PISA 2025 results are not an argument for banning artificial intelligence in schools. On the contrary, they confirm that, as generative AI develops rapidly, one of the most important skills will be the ability to work critically with information, verify it and formulate independent conclusions. This is precisely where the teacher's role remains irreplaceable.

At a glance

Study: PISA 2025

Organiser: OECD (Organisation for Economic Co-operation and Development)

Target group: 15-year-old students

Main assessment areas:

• reading literacy,
• mathematical literacy,
• scientific literacy.

New in 2025

For the first time, the relationship between the use of generative artificial intelligence and student performance was also analysed in greater detail.

Czech Republic

According to the OECD country profile, approximately 70% of Czech students reported that they learn at school to evaluate information created by artificial intelligence. The OECD average is approximately 62.6%.

What is PISA?

Every three years, headlines announcing the results of the international PISA survey appear in the news. Yet many people do not know exactly what this research measures or why its results are considered one of the most important indicators of the quality of education systems.

PISA (Programme for International Student Assessment) is an international study organised by the OECD since 2000. More than eighty countries worldwide, including the Czech Republic, regularly participate. The assessment is taken by 15-year-old students regardless of the type of school they attend.

Unlike ordinary school tests, PISA does not assess how much factual knowledge students have memorised. It focuses primarily on their ability to apply what they have learned to real-life situations. For example, it assesses whether they can understand a specialist text, critically evaluate information, interpret a graph or propose a solution to a practical problem.

This is precisely why PISA results are often used to inform education policy and professional discussions about the future direction of education.

What the PISA 2025 results actually showed

The PISA 2025 results brought several important findings. The continued decline in reading literacy in many OECD countries attracted the greatest media attention, alongside the first more extensive look at the use of generative artificial intelligence in learning. Linking these two topics led to a number of oversimplified interpretations that spread rapidly through public debate.

A closer reading of the results, however, makes it clear that the study itself offers no simple conclusion. Instead, it shows that the relationship between artificial intelligence use and student performance is much more complex.

One of the most discussed findings was that students who used generative artificial intelligence to produce texts for school assignments performed worse in reading literacy, on average, than their peers who did not use AI in this way. This finding became the basis for many media headlines suggesting that artificial intelligence poses a risk to education.

Yet the same study produced another, less frequently cited finding. Students who were also learning the principles of AI literacy – evaluating AI outputs, verifying information, recognising errors and considering the quality of responses – showed slightly better results, particularly in science tasks.

This distinction is crucial.

The results do not suggest that artificial intelligence in itself leads to poorer performance. Rather, they show that how it is used matters. If AI replaces students' own thinking, it can hinder learning. If, on the other hand, it is used as a tool for finding arguments, verifying information or understanding a problem more deeply, it can support the learning process.

From an educational perspective, this is an important message. The discussion should not be about whether to allow or ban AI in schools. The much more important question is how to design teaching so that technology develops independent thinking rather than becoming merely a way to produce a finished solution quickly.

The PISA 2025 results therefore do not provide a simple answer, but confirm a trend many teachers have observed for some time. The mere availability of modern technology does not guarantee better educational outcomes. What matters is how it is used pedagogically, the quality of assignments and the teacher's role in guiding students.

Why media headlines often sounded different from the study itself

After the PISA 2025 results were published, a number of striking headlines appeared in the media. Some suggested that artificial intelligence worsens student performance; others spoke of ‘proof’ that using AI leads to declining reading literacy. Such interpretations are understandable as journalistic shorthand, but they do not reflect what the study actually examined.

Research on this scale generally does not answer whether one phenomenon directly causes another. It primarily describes relationships between variables. In other words, it shows that certain phenomena occur together, not why this happens.

In PISA 2025, students who used generative artificial intelligence to produce school texts were found to perform worse in reading literacy on average. However, this finding does not automatically mean that artificial intelligence was the cause of their poorer results.

Such a conclusion would only be possible if researchers could isolate all the other influences that might affect performance. In an international comparative survey of this kind, that is neither the aim nor methodologically possible.

There are several other explanations that are equally plausible.

Students with lower academic attainment may turn to AI more often precisely because they are seeking help with more difficult tasks. The way school assignments are set may also play a role. If a teacher sets a report that can simply be copied from a language model's output, AI will naturally be used differently than in tasks requiring independent reasoning, comparison of sources or the defence of a viewpoint.

Differences between individual schools, countries or students' social backgrounds cannot be ruled out either. Each of these factors can have a significant effect on performance.

This is precisely why the OECD is very cautious in its interpretation of the results. The report does not say that artificial intelligence in itself causes a decline in reading literacy. It only describes the associations found and points out that interpreting them requires a broader educational context.

For teachers, this is an important reminder for another reason too. At school, we teach students to read critically, distinguish facts from interpretations and verify sources. We should apply the same skill when reading reports on education or new technologies. A striking media headline is not necessarily false, but it often simplifies a more complex reality.

Correlation and causation: a simple example

The difference between correlation and causation can be explained with a simple example.

In the summer months, ice cream sales rise sharply. During the same period, more people swim in outdoor pools, lakes or rivers, and the number of drownings also increases statistically.

It would be wrong to claim that eating ice cream causes drowning.

The two phenomena are indeed related, but their common cause is high temperatures and summer weather. This leads both to higher ice cream sales and to more people being in the water.

The PISA 2025 results need to be approached in the same way. The fact that more frequent use of AI to produce school texts is associated with poorer reading literacy results does not, in itself, tell us what causes that relationship.

This raises a question that is probably more important for Czech schools than the PISA results themselves. If the problem lies not in the existence of artificial intelligence but in how it is used, what conclusions can teachers and school leaders draw?

What the PISA 2025 results mean for Czech schools

The first answer may surprise some people. The PISA results do not support the simple idea that banning artificial intelligence in schools is the solution. Such a step would probably address only the visible symptom, not the cause of the problem. In just a few years, generative AI has become a normal part of students' lives. They use it to find information, do homework and outside school. Schools therefore face a challenge similar to that posed in the past by the internet or smartphones. The question is not whether the technology exists, but whether students learn to use it responsibly and critically. At the same time, it is important to remember that using generative artificial intelligence in schools is not only an educational issue. Schools must also consider legal and ethical aspects, particularly personal data protection, individual service providers' terms and any age restrictions on user registration. These conditions can vary between tools and change over time. Responsible adoption of AI therefore involves not only teaching methodology, but also ongoing monitoring of the legal and contractual conditions of individual services.

It is equally important to think about the design of school assignments. Many traditional tasks originated when the only available sources of information were a textbook or the school library. Today, however, generative AI can produce a coherent text, a book summary or a draft report in seconds. If the only goal of an assignment is to submit a finished text, current technology offers students an easy way to avoid thinking for themselves.

But this does not mean such tasks have lost their purpose. Their purpose changes when the instructions change.

A simple example illustrates the difference. A teacher might assign a report on Charles IV. If the only requirement is to submit a two-page text, almost any language model can produce it in a few minutes. The student need not read a single source or understand the historical context.

Yet the same topic can be assigned differently.

A student could be asked to use generative AI to create a first draft, then verify every claim against two reliable sources, mark inaccuracies or unsupported information, and finally explain in their own words which parts they changed and why. In this case, AI is not a substitute for the student's work, but a tool that opens up opportunities for critical reading, fact-checking and reasoned argument.

In our view, this is where the greatest educational challenge of the coming years lies. Teachers will no longer assess only the final text. Increasingly, they will assess how students work with information, the quality of their reasoning, and their ability to verify facts and use digital tools transparently.

This shift is not actually as radical as it might seem. Czech curriculum documents have emphasised the development of key competencies, working with sources, reading literacy and critical thinking for many years. Generative AI does not change these goals. It simply makes achieving them in practice more urgent.

The PISA 2025 results can therefore also be read as a reminder that quality education does not depend primarily on technology. Thoughtful teaching remains decisive. The same tool can support a deep understanding of the subject in one classroom and merely speed up copying in another. The difference is not made by the algorithm, but by how it is integrated into teaching.

This may be the most important conclusion of the entire study. Artificial intelligence in itself will neither improve nor worsen education. Its real impact will depend above all on whether schools can change their practices so that technology develops thinking rather than replacing it. Instead of creating rules for individual tools, it would be more useful to develop general guidance for the safe and educationally meaningful use of generative artificial intelligence in schools. Such guidance could be developed, for example, under the auspices of the Ministry of Education, Youth and Sports, or in cooperation with expert organisations such as the Czech Association of Artificial Intelligence (ČAUI). Compared with rules aimed at specific applications, broadly formulated principles have one crucial advantage – they are more sustainable over the long term and better able to withstand rapid technological development and the changing terms of individual services.

What do you think?

Do you have experience of introducing AI into teaching?

Have you encountered a similar problem, or perhaps an interesting example of good practice?

Write to us with your perspective. We would be happy to incorporate the most interesting experiences, practical examples or suggestions for further topics into future articles.

👉 Contact: Dvorak.konzultace@gmail.com