15-year-olds are reading, calculating and reasoning worse than ever, thanks to AI, OECD finds

The world's 15-year-olds are reading, calculating and reasoning at the weakest levels recorded since global testing began 25 years ago, according to the OECD's PISA 2025 results released on Tuesday.

The assessment covered more than 760,000 students representing about 33 million 15-year-olds across 91 countries and economies. It found that average performance across OECD countries was the lowest recorded so far in all three core subjects; science, reading and mathematics.

Also read: What should kids study in the age of AI?


Reading recorded the steepest decline. Average scores across OECD countries fell 28 points between 2015 and 2025, and 25 points between 2018 and 2025 alone. Since 20 points is roughly equivalent to one year of schooling for a 15-year-old, the report notes that today's students are reading at a level once expected of students a year younger. Mathematics fell 22 points over the decade, dropping especially sharply after 2018. Science proved comparatively resilient, slipping 7 points over ten years and remaining broadly stable between 2022 and 2025.

Non-OECD participants followed a similar pattern, with reading down 16 points and mathematics down 8 points between 2015 and 2025.

One in five students below the basic bar

Across OECD countries, 20% of 15-year-olds are now low performers in all three subjects, up from 16% in 2022. The report describes these students as lacking the foundational skills needed for further education, employment and everyday life — they struggle to interpret simple data or identify evidence in a graph, find it hard to understand straightforward texts, and have trouble working out the relationship between units and price.

The report points out that these are not students in conflict zones or countries without infrastructure. They live in high- and middle-income OECD countries and have had nine or ten years of formal schooling.

The 'hasty reader'

PISA identifies a specific behavioural shift behind the reading slide. The share of "hasty readers," students who rush through a text and give a quick but incorrect answer, nearly doubled between 2018 and 2025, rising from about 7% to 11%. The share of accurate, fluent readers fell by about 7 percentage points over the same period.

Performance dropped most sharply on tasks built around longer texts, and on higher-order processes such as evaluating information, connecting material across multiple sources and thinking critically about what is read. Tasks that only required locating a single piece of information were the least affected.

The report also finds that scores declined more in the later sections of the test than in the earlier ones, and that students reported spending less effort on the test than in 2022 and 2018.

Wealthier students fell furthest

Socio-economic gaps in performance narrowed between 2022 and 2025, but the report attributes this to declining scores among advantaged students rather than gains among disadvantaged ones. In science, disadvantaged students' performance held steady while advantaged students' scores fell 8 points. In reading, disadvantaged students slipped 4 points against a 20-point drop among advantaged students. In mathematics, advantaged students fell 14 points, while the change among disadvantaged students was not statistically significant.

Socio-economic background nonetheless remains a strong predictor of achievement, accounting for 12% of within-country variation in science performance. Some 37% of disadvantaged students were low performers in science, against 12% of advantaged students.

Digital distraction in the classroom

On average across OECD countries, 28% of students reported that classmates are distracted by digital devices in most or every science lesson. In Beijing-Shanghai-Jiangsu-Zhejiang (China), Japan and Korea, fewer than 10% reported such distraction. Higher distraction was linked to lower science scores in most participating systems.

Also read: AI's left hand begins to take away jobs before its right hand can

Schools that ban cell phones on the premises or set subject-specific guidelines for device use recorded lower distraction levels. Nearly half of students across OECD countries now attend schools where phones are barred on the premises, up from 34% in 2022.

Students reported spending about 1.7 hours a day on digital devices for learning at school and 1.1 hours for leisure while at school. Limited to moderate use for learning was associated with stronger science performance than either no use or very heavy use. Leisure use during school hours beyond one hour a day was consistently associated with lower scores.

AI: widespread, but complicated

Some 86% of students reported using AI chatbots for schoolwork at least occasionally, though near-daily use remains limited. Non-use ranged widely, from around 38% in Japan to under 5% in Vietnam.

The relationship with performance is not straightforward. Students who do not use AI chatbots for specific tasks; summarising texts, drafting written assignments, or preliminary research outperformed those who do. Where AI was used for the broader purpose of "helping me learn", weekly users performed at roughly the same level as non-users, while daily and occasional users scored lower.

Students who frequently used AI and were also taught in class to assess the quality of AI-generated information performed slightly better than their peers. However, such learning opportunities were reported more often by socio-economically advantaged students, which the report flags as a risk of a growing "AI divide".

Falling family support, rising truancy

Students in almost every participating country reported receiving less family support than in 2022, with the decline sharpest among boys and disadvantaged students. Truancy and lateness also ticked up compared with 2018, reversing the improvement recorded in 2022.

On the positive side, students' sense of belonging at school improved in 39 systems since 2022, and fewer students reported feeling lonely at school in 48 systems. Disciplinary climate in science lessons improved in 37 systems between 2015 and 2025, most notably in Korea and Croatia.

Who's on top, and who's climbing

Beijing-Shanghai-Jiangsu-Zhejiang (China) and Singapore were the highest performers across science, mathematics and reading. Estonia, Japan, Korea, Macao (China), Chinese Taipei and the United Kingdom also placed in the top 10 in all three domains.

Several systems bucked the global trend. Among OECD countries, science scores rose in Costa Rica, the Slovak Republic, Türkiye and the United Kingdom since 2022. Among other participants, improvements were recorded in Cambodia, El Salvador, Georgia, Jordan, Mongolia, Montenegro, the Philippines, Saudi Arabia, Thailand, the United Arab Emirates, Uruguay and Uzbekistan.

Türkiye, the UAE, Montenegro, Qatar and the Slovak Republic managed to cut their share of low performers while raising their share of top performers.

Spending is not destiny

National wealth accounts for about half the variation in performance across countries, but the report stresses that spending levels alone do not determine outcomes.

Above roughly USD 85,000 in cumulative spending per student between ages 6 and 15, the link between money and results begins to break down.

Luxembourg spends USD 288,521 per student, by far the most of any participant, but posted lower average science scores than Ireland, France and Italy, each of which spends less than half that amount.