FAQ — G3O

The project

What is G3O?: G3O (Global Government GenAI Observatory) is a cross-national, cross-institutional measurement project that documents the adoption of generative AI by public institutions worldwide. It produces a versioned, open-source panel dataset with transparent validation.
What counts as public GenAI usage?: G3O documents any use of generative AI by a government entity: public statements, policies, pilot programs, active deployments, chatbots, internal tools, procurement decisions, and governance frameworks. We distinguish between stated intent and verified deployment.

How do I use the registry?: Search for an institution by name, or use the browse-by controls to narrow by country, branch, or government level. Opening an institution shows its profile, including each coded finding and the sources behind it. Selecting two or more institutions puts them side by side.
How can researchers access the data?: Versioned public releases and a query API are in preparation. Until they land, the registry can be searched here without an account, and researchers who need an extract can make a bulk-access request.
Can I use G3O data in my own research?: Yes. G3O is built to be an open research resource and academic use is welcome; we ask only that you cite the project. For access before a public release, write to us about a collaboration.
How can I contribute data?: We welcome contributions from researchers, practitioners, and the public. Choose “Contributing data” on the contact form, or write to us. A structured submission workflow and guidelines are still to come; for now, contributions are handled by email.
How do I report an error?: On every institution profile you can tell us that something looks wrong, and the contact form has an option for reporting an error. We investigate all reports. A wrong value with a source attached is more useful to us than a missing one.

How is data collected?: G3O uses a multi-method approach: (1) structured, AI-augmented web searches using institution-specific queries, and (2) parsing and extraction of structured fields via LLM API calls held to a fixed, machine-checked output structure.
What are the limits of the current data?: Collection is still under way, and the current dataset has known limitations: geographic coverage is uneven (with more data from Europe and North America), many entries are based on public web sources rather than verified institutional records, and the dataset is updated periodically rather than in real time. Where the pipeline can miss evidence, and where a record can be wrong, is set out in full on the collection-pipeline page.