AI in official statistics: planning, AI-ready dissemination and measuring adoption

The Statistics Centre – Abu Dhabi (SCAD) presented its AI agent Suhail at the KEZAD 2026 Forum, held at the Abu Dhabi National Exhibition Centre (ADNEC). WAM and Emarat Al Youm reported the news on 15 October 2026. Suhail is described as a 'statistical portfolio assistant'. It helps government entities and industrial companies turn their priorities and operational requirements into specific statistical needs. Specialised quality teams supervise its work.

Director General Abdullah Ghareeb Al Qamzi said that the centre links statistical planning to user needs through the responsible use of AI. Suhail is a tool for identifying statistical needs. It does not produce statistics.

At the same forum, the centre presented two other initiatives. 'Data for the Public Good' supports secure and responsible data exchange between government and the private sector, while preserving confidentiality. The 'Bayan' digital platform offers interactive dashboards and official reports on economic performance, population and the labour market. The report does not give any figures.

On 10 March 2026, the IMF published Departmental Paper No. 26/04, 'StatGPT: AI for Official Statistics'. The authors are James Tebrake, El Bachir Boukherouaa, Jeff Danforth and Nivashini Harikrishnan. The work is an initiative of the IMF Statistics Department.

The paper finds that general generative AI tools such as ChatGPT and Gemini are good at summarising text, but perform poorly when asked for official statistics. They often give figures that are 'dangerously reasonable but incorrect'. StatGPT takes a different approach. It uses large language models not to generate statistics, but to generate structured queries to the APIs of official statistical agencies. This way, the system retrieves the exact published figure.

The paper also sets out a roadmap for AI-ready official statistics. It recommends open data access through APIs, richer metadata standards and stronger data governance, so that official statistics remain authoritative and trusted. The paper can be downloaded free of charge as a PDF from the IMF website.

On 6 February 2026, India's National Statistics Office (MoSPI) launched a beta Model Context Protocol (MCP) server for its eSankhyiki portal. With the server, users can connect AI tools and assistants directly to official datasets, so they do not need to download files and reformat them.

Seven datasets were available at launch: the Periodic Labour Force Survey, the Consumer Price Index, the Annual Survey of Industries, the Index of Industrial Production, National Accounts Statistics, the Wholesale Price Index and Environmental Statistics.

The service is still in beta. It is an example of an NSO making its official statistics 'AI-ready' at the dissemination stage, so that AI assistants can use the official data directly.

PARIS21 Academy is the free online learning platform of PARIS21 for the statistical community. It offers self-paced interactive courses and skill-building modules on topics such as data planning and data communication.

The courses are available at any time. Because they are self-paced, individual staff members and whole teams can fit their learning around production schedules. This makes the platform a flexible option for NSOs that want to build staff skills alongside regular work.

On 2 October 2026, the U.S. Bureau of Economic Analysis (BEA) published a research spotlight in the Survey of Current Business, titled 'AI Utilization and Changes in Economic Performance'. It is based on a working paper by Tina Highfill and Jon D. Samuels of BEA and Christos A. Makridis (Gallup, Arizona State University and Stanford University).

The study combines three sources: worker-reported AI use from the Gallup Panel, employer-reported AI adoption from the Census Bureau's Business Trends and Outlook Survey (BTOS), and LEHD and BEA data on employment, earnings and real output by state and industry. The share of workers reporting any AI use rose from roughly 20 percent in mid-2023 to nearly 50 percent by early 2026. Frequent use rose from about 10 percent to more than 25 percent. Over a comparable period, employer-reported adoption rose from about 4 percent to 18 percent.

Worker-based estimates are higher than employer-based ones by roughly 15 to 32 percentage points. However, across industries the two measures are strongly correlated (0.93). State-industry groups with higher worker AI use had stronger growth in real output after 2020. Differences in employment were generally positive but less precisely estimated. BEA notes that reverse causality is possible and calls the work an early step in a broader measurement project.