OECD warns AI could affect trust in official statistics
Official statistics need stronger traceability, structure and stewardship as AI reshapes data access.
The OECD has warned that generative AI is changing how people access official statistics, creating new challenges for data quality, context and public trust.
In a blog published by the OECD Centre on Well-being, Inclusion, Sustainability and Equal Opportunity, Romina Boarini and Cameroon Tiati A Biscene argue that citizens, journalists and policymakers increasingly obtain official statistics through chatbots and AI assistants rather than directly from national statistical institutes or government websites.
According to the OECD, this lengthens the chain between data producers and users. As statistics pass through AI systems and other digital intermediaries, important context, including reference periods, revision notes, methodological caveats and source attribution, can be lost.
The blog notes that AI can make official statistics easier to discover and understand, particularly for non-specialist audiences. However, it warns that AI-generated summaries often obscure how information has been selected, interpreted and simplified before reaching users.
The OECD also highlights several data quality risks. AI systems may fail to recognise when official statistics have been revised, corrected or withdrawn, while retrieval quality depends heavily on how well statistical information is structured and machine-readable. Poorly organised data can therefore increase the risk of inaccurate or misleading outputs.
The report also raises concerns about representativeness. General-purpose AI systems are trained on vast amounts of online content, but abundance does not guarantee representative data. As a result, AI-generated or synthetic representations of populations may fail to reflect real-world conditions accurately.
Access is another concern. Although official statistics remain publicly available, meaningful access may increasingly depend on private AI assistants, paid interfaces and concentrated digital infrastructure, potentially making a public good less accessible in practice.
The OECD argues that national statistical institutes may need to expand their role by making datasets more structured and machine-readable, monitoring how statistics are reformulated by AI systems, developing standards for unofficial data sources and preserving the institutional independence that underpins public trust.
Why does it matter?
The OECD’s warning highlights that official statistics can remain accurate at source yet become misleading once AI systems summarise, simplify or detach them from their original context. As more people rely on AI assistants rather than official websites, preserving context and source attribution will become increasingly important for maintaining trust in public data.
The findings also suggest that national statistical institutes will need to adapt to an AI-mediated information environment by designing datasets not only for human users but also for AI systems that increasingly act as intermediaries between governments and the public.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!