An OECD blog post by Andreas Schleicher, citing PISA survey data, reports widespread student use of generative AI for schoolwork and lower average science scores among students who use it for some tasks. The author stresses the findings are correlational, based on self-reported use, and vary by country.
How this is computed
- Key claims supported: 5 of 6
- Type of source: Institution on its own work
- Independent confirmation: None
- Rule: Without independent confirmation at most level 3
Rates the source, not the topic and not our analysis.
Why it matters
Our reading: PISA offers a large cross-country view of how secondary-school students report using AI, but the post illustrates associations, not effects. Score gaps could reflect which students turn to AI rather than what AI does to learning, so the data supports caution more than firm policy conclusions.
What the source says
- Conditionally supported The OECD says students who use AI chatbots for tasks such as summarising, preliminary research or drafting score about 20 points lower in science on average than non-users, which it equates to roughly one year of schooling; the author notes struggling students may rely more on AI.
- Conditionally supported According to the OECD, students who report using AI chatbots to help them learn once or twice a week perform at a similar level to non-users on average across the OECD.
- Conditionally supported The OECD reports that in Canada and New Zealand, students who used chatbots to draft writing texts fairly regularly or daily scored 29 and 36 points lower; PISA sampling standards were not met in both countries.
- Conditionally supported The OECD reports the opposite pattern in Thailand and Macao (China), where weekly or daily AI drafting users scored 24 and 21 points higher than classmates.
- Conditionally supported Per the OECD, more than 40% of students in several systems, including Singapore, Saudi Arabia and Türkiye, say they use AI to draft writing texts at least weekly, reaching 52% in Viet Nam.
- Vendor claim only Citing its own Digital Education Outlook, the OECD says educational GenAI tools can lead to sustained learning gains while general-purpose tools often do not support deep learning; the post does not detail the underlying studies.
The key caveat
All figures come from self-reported, cross-sectional survey data presented by the OECD about its own programme; none of the score differences is shown to be caused by AI use.
Open questions
- Do the score gaps persist after accounting for prior achievement and student background?
- What explains the opposite associations in Thailand and Macao (China) versus Canada and New Zealand?
- Which studies underpin the Digital Education Outlook’s distinction between educational and general-purpose AI tools?
What’s next
Watch for the published PISA 2025 volumes and country tables, which would allow the cited figures to be checked.
Source: OECD Education and Skills Today, 13 October 2026: Generative AI and student learning: What the PISA data reveal
Produced by Zeitgeist’s AI agents (research, editing, standards check) with Claude. No human review.