Students who use AI for specific schoolwork tasks such as summarizing texts, drafting or conducting research score lower in science than those who do not, by around 20 points on average, a gap the OECD equates to roughly one year of schooling.
At 66%, Singapore’s weekly use rate runs 20 percentage points above the OECD average, and only 4% of its students reported never or almost never using AI chatbots for any of the schoolwork tasks PISA examined, against an OECD average of 14%.
Some 46% of Singapore students use AI to conduct preliminary research on a new topic against an OECD average of 31%, 45% to draft text for writing assignments against 29%, and 41% to summarize a text they had to read against 30%.
TechNode Global reported the widest gap was for summarizing reading, where daily users scored nearly 30 points below non-users.
Bloomberg put the drafting comparison at 509 points for students who never or almost never use AI to write assignments against 481 for daily users, a 28-point difference adjusted for socio-economic status. The OECD describes these as associations, and notes that weaker students may be more likely to reach for AI in the first place.
Weekly users of AI for general purposes performed similarly to non-users, and students who used a chatbot about weekly specifically to help them learn posted the highest science scores of any group at around 500 points.
Andreas Schleicher, the OECD’s director for education and skills, told The Straits Times that AI separates task performance from learning, letting students “do wonderful things without learning anything”. He argued that AI’s fluency and persuasiveness make it feel like common sense, and that common sense is no longer sufficient at a point when a shared, verifiable reality needs reinforcing against intuition.
Where Singapore’s scores landed
Singapore’s mean science score was 560, highest in the world after China, and against an OECD average of 482, with low performers at 9.9% against 25.7%. Some 37% of its students were top performers in mathematics, reaching Level 5 or 6, against an OECD average of 8%, and 86% attained Level 2 or higher in reading against 69%.
The highest performers across science, mathematics and reading were the Chinese jurisdictions of Beijing, Shanghai, Jiangsu and Zhejiang, and Singapore. Estonia, Japan, Korea, Macao (China), Chinese Taipei and the United Kingdom were also in the top 10.
In Singapore, 29% of students reported that peers are often distracted by digital devices during most or all science lessons, against an OECD average of 28%. That distraction cost Singapore students 6 points in science after adjusting for socio-economic profile, compared with 11 points across the OECD.
In 2025, 43% of Singapore students attended schools banning cellphone use on the premises, up from 16% in 2022 and still short of the OECD average of 49%.
Across the OECD, students in schools with a ban are about 13% less likely to report digital distraction.
The Straits Times reported that 55% of Singapore students said a parent or guardian asked at least weekly about their school day and 37% had weekly discussions about problems at school, figures Schleicher tied to a concern that families are becoming clients of education rather than co-producers.
Singapore’s Ministry of Education acknowledged the finding and pointed parents to its Parenting for Wellness resources.
Some 23% of Singapore students reported no exercise outside school against an OECD average of 17%, an improvement on 2022, when the ministry reported 29% not exercising at all after school against an OECD average of 20%.
PISA 2025 tested more than 760,000 students across 91 economies with science as the focal domain, the largest cycle in the programme’s history and the first since generative AI reached mainstream use.