New Delhi: The latest global education assessment results show that the world’s top-performing school systems are taking very different approaches to AI in education.
China’s highest-performing jurisdictions—Beijing, Shanghai, Jiangsu and Zhejiang—have topped the latest Programme for International Student Assessment (PISA) in mathematics and science, yet their students report relatively restrained use of AI for learning.
Singapore, the second top performer, takes a markedly different approach: Its students report much higher AI use, while also receiving substantially more instruction on how to assess AI-generated information.
The findings come from the 2025 PISA, which is conducted every three years by the Organisation for Economic Co-operation and Development (OECD), a group of 38 member countries that works with governments to improve economic and social policies. The latest report was released Tuesday.
PISA assesses the knowledge and skills of 15-year-old students in reading, mathematics and science, and compares how education systems are performing across participating countries and economies. Countries such as China, Singapore, Japan, South Korea, Finland, the UK and the US participate in the assessment.
India does not currently participate in PISA. It last took part in the 2009 assessment. India was placed 72nd among the 74 participating countries. However, only two states Tamil Nadu and Himachal Pradesh, were represented. Although India has not officially cited any reason, there were concerns that PISA’s competency-based format was unfamiliar to Indian students.
According to the latest PISA report, students in Beijing, Shanghai, Jiangsu and Zhejiang—grouped by the OECD as B-S-J-Z (China)—scored 612 in mathematics and 597 in science, the highest scores among all participating education systems. The region ranked second in reading, with a score of 527, after Singapore.
Singapore, meanwhile, scored 563 in mathematics and 560 in science, placing it second in both, while topping the reading table with a score of 535.
However, the contrast becomes striking when the results are viewed through the lens of AI. Although the PISA findings do not show that either approach produces better scores, they raise a more important question for schools: How to find a right balance?
China’s top performers not the biggest AI users
When it comes to AI, the Chinese jurisdictions do not report the highest levels of student use.
About 41 percent students in B-S-J-Z (China) said they use AI chatbots at least weekly to help them learn, compared with 46 percent across OECD countries. The proportion was also lower than in several high-performing developed education systems.
China’s relatively restrained student use of AI does not reflect a lack of government push. Beijing has issued AI-literacy guidance for schools since 2024, followed by rules on students’ use of generative AI in 2025. In April 2026, five government departments launched an ‘AI+Education’ action plan, stressing appropriate use, critical thinking and problem-solving.
However, the PISA data offers no evidence that lower AI use itself explains China’s higher scores. In fact, OECD puts it more directly: “AI use is not a prerequisite for success, at least at the time of PISA 2025.”
Instead, the report finds a more complicated relationship between AI use and performance. Across systems, students who do not use AI for particular schoolwork tasks often perform better than those who do, although “moderate” use for some purposes can be associated with better outcomes than either limited or intensive use.
The OECD stresses that these are associations, not evidence that AI use—or avoiding it—causes higher or lower achievement.
What are students actually using AI for?
The gap between China and the OECD average becomes more pronounced when looking at specific schoolwork tasks.
While 31 percent students across the OECD said they use AI to conduct preliminary research on a new topic, only 17 percent in B-S-J-Z (China) reported doing so. The corresponding figure was 46 percent in Singapore, 44 percent in Korea and 35 percent in Australia.
Similarly, 30 percent OECD students said they use AI to summarise a text they had to read, compared with 19 percent in B-S-J-Z (China). The figure was 41 percent in Singapore, 37 percent in Korea and 33 percent in Australia.
For drafting writing assignments, 29 percent OECD students reported using AI, against 21 percent in B-S-J-Z. Singapore was again much higher at 45 percent, followed by Korea at 37 percent and Germany at 36 percent.
The figures suggest that the gap is not simply about whether students have access to AI, but about how frequently they are using it for particular academic tasks.
The OECD cautions: “Simply expanding access to AI tools is unlikely to produce benefits without parallel investments in guidance on appropriate and effective use.”
That distinction is important in the Chinese case. Its highest-performing jurisdictions are not disconnected from digital technology; rather, students report less frequent use of AI for tasks such as researching, summarising and drafting.
China also reports much lower digital distraction
According to the report, only 5 percent students in B-S-J-Z said their peers were distracted by digital devices during most or every science lesson, compared with 28 percent across the OECD.
The Chinese jurisdictions also have stringent restrictions on mobile phones. About 83 percent students said they attended schools where cellphone use was not allowed on school premises, compared with about 50 percent across the OECD.
Students in B-S-J-Z also reported spending less time on digital devices at school for learning—about 1.2 hours a day, compared with 1.7 hours across the OECD—and less time using devices for leisure.
But again, it cautioned that stricter phone rules or lower digital use produced better PISA scores.
Singapore model: High performance with high AI use
The sharpest contrast comes from Singapore, where AI use for learning is substantially higher than both the OECD average and B-S-J-Z (China).
About 66 percent students in Singapore said they use AI chatbots at least weekly to help them learn, compared with 46 percent across the OECD and 41 percent in B-S-J-Z (China).
Singapore also reports particularly high levels of AI-related learning in schools. 81 percent students said they learn in school how to assess the quality of AI-generated information, compared with 63 percent across the OECD.
This makes Singapore an important counterpoint to China: Both are among the strongest-performing education systems in PISA, but their students report markedly different levels of AI use.
Singapore’s students are not simply using AI more frequently; the system also reports substantially greater exposure to teaching how to evaluate AI-generated information.
So, does more AI mean better or poorer learning?
The PISA findings therefore do not show that more AI means more learning. They point instead to the importance of how AI is used and what students are taught alongside it. It states that students who frequently used AI to help them learn and were also given opportunities in school to assess AI-generated information tended to slightly outperform their peers. But the OECD cautions that “these results cannot determine causality”.
What the findings do underscore, the OECD says, is the “potential of targeted instruction” to strengthen students’ critical engagement with AI.
That may be the clearest lesson from the two top-performing systems examined here. China shows that exceptional performance does not require heavy student use of AI, while Singapore shows that high AI use can coexist with high performance when students are also taught to evaluate AI-generated information.
Neither model proves that AI use—or limiting it—causes better academic performance.
The report states that “digital readiness does not begin with devices, it begins with helping young people cultivate the ability to concentrate, interpret, synthesise and reason—capacities that are deeply tied to the habits of deep reading. This is not a call to turn back the digital clock, but to find balance.”
(Edited by Viny Mishra)
Also read: Students use AI, parents panic—how Indian schools are finding ways to live with ChatGPT