Reading levels among teenagers worldwide fell to its lowest level ever recorded in the history of the PISA international assessment. A new OECD study covered more than 760,000 15-year-old pupils from 91 countries and economies — this is the largest-scale assessment in the programme’s history. The average score for reading has fallen from its peak 501 points in 2012 to 466 points in 2025. According to the OECD, the average 15-year-old pupil today has a reading level that was previously typical of a child about a year younger.
- Reading standards have fallen to their lowest level in the history of PISA.
- The study covered 760,000 teenagers in 91 countries.
- Time spent on social media has increased, whilst reading for pleasure is on the decline.
- 46% pupils regularly use AI for their studies.
- In Ukraine, 54% schoolchildren use AI at least once a week.
Over the past ten years, teenagers have lost around one and a half years’ worth of academic progress
Pros. data OECD: the average reading score across the organisation’s member countries fell by 28 points between 2015 and 2025. This is equivalent to approximately a year and a half of study.
Mathematics results for the same period fell by 22 points — equivalent to more than one academic year. Overall, average scores in reading and mathematics were the lowest since PISA began.
Reuters also paints a long-term picture: maths has fallen from a high of 502 to 469 points, whilst the natural sciences rose from 506 to 486 points.
Another worrying indicator is — One in five 15-year-old pupils in OECD countries now performs poorly in reading, maths and science all at the same time. In 2022, there were 16% of these.
The OECD has turned its attention to smartphones and social media
One of the most noticeable changes concerns the fact that, how teenagers consume information.
Secretary-General of the OECD Matthias Korman stated that an increase in screen time and a decline in reading for pleasure are occurring in parallel with a deterioration in reading outcomes. At the same time, the OECD does not claim that smartphones themselves are the direct cause of the decline — the report refers to a statistical correlation and changes in pupils’ behaviour.
On average, teenagers in OECD countries spend around six hours on a weekday, taking into account study and leisure activities.
Approximately 3.4 hours a day is spent specifically on entertainment — social media, videos, games and other content.
Compared with 2022, the proportion of pupils spending more than an hour a day on social media has also increased.
Children are increasingly «skimming» texts rather than reading them carefully
The OECD draws particular attention to the phenomenon hasty reading — reading in a hurry.
This refers to a situation where a pupil quickly scans the text and immediately chooses an answer, but gets it wrong. The proportion of such cases has almost doubled since 2018 and reached around 9% in 2025.
At the same time, it is precisely these deep reading skills that are becoming increasingly important in the age of AI: the ability to verify information, compare several sources, spot contradictions and critically evaluate what one has read.
The PISA data also show that pupils who read more often paper books, on average, perform better than those who mainly read on digital devices or hardly read any books at all. The OECD cautions that this does not prove a direct causal link either.
Almost half of school pupils are already using AI for learning
PISA 2025 provides, for the first time, a comprehensive overview of school pupils’ use of generative artificial intelligence.
On average, in OECD countries 46% teenagers use AI chatbots at least once a week, …to help themselves with their studies. Around one in five uses such tools almost every day.
Pupils apply ChatGPT- similar services for:
- searching for initial information;
- writing academic papers;
- paraphrasing and summarising texts;
- explaining complex topics;
- preparation for lessons.
And here, the OECD identified a telling pattern.
Pupils who used AI to writing texts, summarising what has been read and researching materials, on average, scored approximately 20 marks fewer in the natural sciences — the difference is roughly equivalent to one year’s study.
But that does not mean that the use of AI automatically makes learning worse.
AI can be both a hindrance and a help
The OECD emphasises that the outcome depends on whether, what exactly the child uses artificial intelligence for.
Pupils who simply ask a chatbot to write an essay or summarise a text for them tend to achieve poorer results.
However, amongst those who use AI as a learning tool — for example, by asking it to explain a topic — moderate use is not associated with the same deterioration in outcomes.
What is more, schoolchildren who were taught at school to check the quality and accuracy of AI responses, showed slightly better results than active users who had not received such training.
In essence, the OECD’s conclusion is that AI can be a useful aid, but should not replace the thought process itself.
More than five hours of digital learning — results start to dip
The study also showed that digital devices are not a problem in themselves.
Moderate use of laptops, tablets and other devices for learning at school is linked to better results in science.
However, after about five hours a day The trend is changing — the results are starting to get worse.
An even greater impact is caused by the use of smartphones for entertainment whilst in class. After just about an hour of such use, performance in science subjects gradually declines.
More than one in four schoolchildren reported that pupils are distracted by digital devices during most or all science lessons.
And what did PISA 2025 reveal in Ukraine?
Ukraine also took part in the study, although, due to the war, the assessment was carried out in 17 out of 27 regions.
The following took part in the testing 7,105 Ukrainian pupils from 294 schools, representing approximately 256,000 15-year-olds.
The following countries achieved the minimum PISA reading standard: 55% Ukrainian schoolchildren, whilst the OECD average stands at 69%.
The following have achieved the minimum standard in mathematics 54% compared with 65% in the OECD, in the natural sciences — 65% versus 74%. Overall, the results of Ukrainian pupils in 2025 remained roughly at the same level as in 2022.
The difference is particularly noticeable in the use of AI.
54% Ukrainian teenagers use AI chatbots for learning at least once a week, whilst the OECD average stands at 46%.
In addition:
- 42% Ukrainian schoolchildren use AI to paraphrase texts;
- 43% — for writing academic papers;
- 37% — for an initial search for information;
- only 8% They stated that they hardly ever use AI for educational purposes.
Who achieved the best results?
The education systems of East Asia once again achieved the highest scores in PISA 2025.
Among the frontrunners are the Chinese regions that took part in the trials, Singapore, Japan, South Korea, Macau and Taiwan.
Among countries outside East Asia, the following made it into the top ten in reading, maths and science: Estonia and the United Kingdom.
At the same time, the OECD emphasises that an overall decline is not inevitable. A number of countries have managed to improve certain outcomes even against the backdrop of a global downturn.
What do the PISA results mean?
The study does not suggest that smartphones, tablets or AI should be completely abandoned in education.
The OECD also did not recommend a blanket ban on smartphones in schools. However, the results show that the problem arises when technology begins to replace concentration, careful reading and independent problem-solving.
In its findings, the OECD emphasises that technology and artificial intelligence can enhance learning, if they are used in a targeted manner and do not serve as a substitute for attention, effort and understanding.
And it is precisely the ability to concentrate for long periods, to read critically and to verify information that may become one of the most important skills for a generation growing up alongside generative AI.
