‹ News Blitz Social Settled

PISA study finds no simple link between AI use and better scores

English हिंदी ಕನ್ನಡ മലയാളം தமிழ் తెలుగు

First brief 14 Sep, 9:32 pm IST Updated 14 Sep, 9:32 pm IST 0 developments 3 min read

File: School laptops — Kaden Pitt / U.S. Army · Public domain

Where it stands

An international school study suggests that access to an AI chatbot is not enough to ensure better learning. Students use these tools for different purposes, from asking for explanations to drafting assignments and summarising reading material. PISA 2025 compared those reported habits with students’ science performance. Non-users generally scored higher than students using AI for specific schoolwork tasks. But students using AI weekly for the broader purpose of learning performed similarly to non-users after differences in socioeconomic background were considered. The distinction matters because these patterns do not prove that AI caused a score to rise or fall. Students who choose the tools may already differ in their needs, opportunities and learning habits. The study also found an association between stronger performance and opportunities to assess AI-generated information at school among frequent users. Such opportunities were less common for disadvantaged students. Published by the OECD on 8 September 2026, the findings invite closer examination of school guidance as well as access to AI. India is not covered in these results, so they should not be presented as measurements of Indian students.

Background

A student can finish an assignment with help from a chatbot without necessarily understanding every step in the answer. Understanding means being able to explain an idea, evaluate evidence and apply it elsewhere, beyond the finished assignment. This is why the effect of AI in education cannot be judged only by whether a tool saves time or produces fluent text. Consider a hypothetical student reading about the water cycle. Asking a chatbot to summarise the passage may shorten the reading task. Asking why cooling water vapour forms droplets, then checking the explanation against the lesson, involves a different kind of engagement. These are examples of different uses, not cases observed in the study or proof that one method always works. They show why researchers need to ask what students do with AI, as well as how often they use it. PISA, the Programme for International Student Assessment, examines how students around age 15 apply reading, mathematics and science knowledge. Alongside the tests, questionnaires gather information about their backgrounds and learning environments. Researchers can then compare test performance with reported habits, including AI use. But the students were not randomly assigned to use or avoid AI. Differences between the groups may therefore reflect earlier achievement, support at home, access to technology or other influences. Statistical adjustments can account for measured background differences, but they cannot establish that all other influences have been removed. The findings are consequently useful for identifying questions schools should investigate, rather than prescribing a universal amount of AI use. Guidance on checking an answer’s evidence and reasoning matters because a confident explanation still needs to be assessed before a learner relies on it.

How it developed

8 September 2026: PISA results released

How it started
The assessment separates AI access from patterns of learning

The OECD released the first volume of PISA 2025 results on 8 September 2026. More than 760,000 students participated across 91 countries and economies. The assessment focused on science, alongside reading, mathematics and learning in the digital world. Questionnaires asked students about AI use for summarising assigned texts, researching new topics, drafting assignments and helping them learn. The results show why those purposes should be examined separately. Students not using AI generally scored higher in science than those using it for specific schoolwork tasks. Weekly use for learning showed a different pattern. After accounting for socioeconomic background, students who used AI weekly to help them learn performed similarly to non-users. Among frequent users, opportunities at school to assess AI-generated information were associated with slightly stronger performance. These learning opportunities were less often reported by disadvantaged students. These comparisons cannot establish cause and effect. They nevertheless point to unequal opportunities for students to receive help in understanding and evaluating AI output.

OECD OECD Business Standard

Why it matters for UPSC

GS2GS3

For GS2, connect learning outcomes with the quality and fairness of educational support. For GS3, examine responsible uses of artificial intelligence and the digital divide. Distinguish an observed association from a demonstrated cause before drawing a policy conclusion from a survey.

Key terms

PISAThe Programme for International Student Assessment compares how students around age 15 apply knowledge in reading, mathematics and science. Tests are accompanied by questionnaires about learning and background. It examines more than recalled facts from a particular textbook. Results describe participating education systems, not every country automatically.

OECDThe Organisation for Economic Co-operation and Development is an international organisation that studies economic and social policies. It runs PISA with participating countries and economies, including some outside its membership. An average for OECD countries is therefore not automatically an average for every system participating in PISA.

AI chatbotA computer system that generates conversational responses to prompts. Students may use it to explain a topic, suggest text or summarise material. A fluent response can still contain errors or omit important qualifications. Using such a tool and understanding its answer are separate steps in learning.

Correlation and causationCorrelation means two things vary together in the observed data. Causation means a change in one produces a change in the other. A score difference between AI users and non-users does not, by itself, prove an effect of AI. Other differences between the students may help explain the result.

Socioeconomic backgroundThe social and economic circumstances surrounding a student, including family resources and educational opportunities. These circumstances can affect access to devices, learning support and school performance. Researchers consider them when comparing groups, so that differences are not immediately attributed to the technology being studied.

Statistical adjustmentA method for comparing outcomes after taking specified measured differences into account. Here, researchers consider socioeconomic background when comparing science scores. Adjustment can improve a comparison, but it does not remove every possible influence or turn a survey into a randomised experiment.

AI literacyThe ability to understand, use and critically assess AI tools and their outputs. For a student, this includes checking whether an answer has sound evidence, fits the question and contains errors. It involves judgement, not merely knowing how to type a prompt or copy a response.

Digital divideUnequal opportunities to access and benefit from digital technology. The gap can involve devices and internet connections, but also skills, guidance and support. In education, giving students a tool does not ensure that everyone receives equal help in using it effectively.

Sources (3)

OECD · official · PISA 2025 Results, Volume I8 Sep, 1:00 pm
OECD · official · PISA: student school life and beyond8 Sep, 1:00 pm
Business Standard · AI helps with work, but more use does not mean better scores8 Sep, 1:00 pm