Diia.AI: Ukraine's national AI assistant for government services

The problem

Governments around the world struggle with the same tension: public services are often built around administrative logic rather than the citizens' experience of using them. In Ukraine, accessing a government service online typically meant knowing which of the many portals to visit, determining which specific service applied to their situation, and completing multi-step forms manually. Many procedures still require an in-person visit.

Ukraine had already made significant progress in changing this. The Diia platform brought over 150 government services into a single app and website, from digital identification to administrative certificates. Nearly 23 million people used it. But having services available online is not the same as making them easy to use. Citizens still had to find the right service, understand which one applied to their situation, and navigate each step themselves.

For many people, that remained difficult. Older citizens, people less confident with digital tools, and those accessing services from unfamiliar locations could find even a well-designed system hard to follow. Complex and opaque processes had also, as the European Commission noted, provided openings for corruption. The Diia platform had brought services online. The remaining challenge was making them simple to use.

The solution

Ukraine's Ministry of Digital Transformation built Diia.AI together with its in-house technology company Diia and two industry partners, KitSoft and Google. It is an AI assistant on the Diia platform, built on a combination of Google's Gemini 2.0 Flash and Ukraine's own secure government systems, that completes government services for citizens inside a chat conversation.

A citizen opens a chat window and describes what they need in plain, everyday language, such as requesting a document or checking an entitlement. The assistant connects to the relevant government databases, retrieves the citizen's information, and completes the process within the conversation. What previously required locating the right form, filling it in manually, and submitting it through the correct channel now happens through a short exchange of messages. Unlike a conventional government chatbot, which answers questions or points users to the right page, Diia.AI is connected to live government systems and can complete transactions on the citizen's behalf.

By the Ministry's account, the assistant works at three levels. The first is guidance: it can walk a citizen through how to reach any service on the portal and explain the ministry's programmes, answering questions such as how to sell a car through Diia or how to choose a business grant on Diia.Business. The second is matching a service to a person's situation: someone whose home has been damaged can be pointed to the eRestoration scheme, told how to submit photographs, and shown how to send an application to the relevant commission. The third, and the most recent, is carrying out the service in the chat itself. At launch this covered a single service, an income certificate, with more to be added as the system develops, working toward the point where the assistant becomes one entry point to the services on the portal.

A central design question was how to use a powerful AI tool without exposing sensitive personal data. The system uses a privacy-by-design approach built as a four-stage process: before any request reaches the AI model, all identifying information is swapped for neutral placeholders. The AI processes the request without ever seeing who the citizen is, and a secure government layer then matches the response back to the right person. The result is that capable commercial AI tools can be used for their language abilities while all sensitive data stays within government-controlled infrastructure. The system also includes protections against attempts to manipulate the AI into producing unintended or harmful responses, developed specifically for the Ukrainian language.

Diia.AI sits within Diia.Engine, Ukraine's national digital infrastructure, which connects services run by central government bodies through to local authorities. This existing foundation meant the assistant could draw on live government data from the outset, rather than working from a fixed set of pre-written answers.

Rather than trying to cover every service at once, the team looked at the twenty most-requested services on the platform and began with those that drew the highest volume of enquiries. Each of these services was redesigned around how a citizen might naturally ask for it in conversation. The assistant was tested through pilot programmes before being opened to the public in a testing phase, with feedback from real users shaping ongoing improvements.

The impact

1. Faster, simpler service delivery, with early results to match

The most immediate measure of whether a government service is working is whether citizens use it and find it useful. Early results from Diia.AI are encouraging on both counts. User satisfaction scores from the initial rollout sit above 80%. That reflects ease of use, and also a more basic change in what citizens are asked to do: where they once had to navigate a system, they now describe a need and have it handled.

The time savings are substantial. Requests that once required finding the right portal, identifying the correct form, and working through a multi-step submission are now handled within the chat itself, with a projected saving of more than one million citizen-hours a year. On the government side, a projected 30% reduction in enquiries that need a staff member to handle would let staff time now taken up by routine, high-volume requests go instead to the cases that genuinely need human judgement.

These figures come from the early adoption phase and will need to be confirmed as usage scales, but they are consistent with what the system was designed to achieve: fewer steps, less time, and fewer points at which a citizen needs to ask for help.

2. Government becomes easier to use for those who found it hardest

The shift to a conversational interface has a different significance depending on who the citizen is. For a digitally confident user, it is a convenience. For someone who found government websites difficult to navigate, it is more substantial.

The previous system rewarded familiarity, knowing which portal to use, understanding administrative terminology, and having the time and confidence to work through a multi-step process. Diia.AI removes those prerequisites. A citizen does not need to know how the government has organised its services; they need only describe what they need. For citizens who are older or less confident with digital tools, removing the need to know which portal to use or which form to find lowers a real barrier to using public services on their own. A voice capability now being added to Diia, announced with the company ElevenLabs, is aimed at these users in particular, including people who find typing or screens difficult.

3. An internationally recognised first in government AI

In November 2025, the World Record Book recognised Diia.AI as the world's first national AI assistant for government services at the WINWIN Summit in Kyiv. The same month, it received the Government-to-Citizens Best Cases Award at the European Commission's 2025 SEMIC Conference in Copenhagen.

Lessons learned

• An established digital platform enables the quick addition of an AI layer. Diia.AI did not require government systems to be rebuilt. It was connected to services already running on Ukraine's national low-code platform, which had been built up over the preceding years. Having that groundwork already in place is what allowed the assistant to be launched as quickly as it was.

• Citizen data can be kept out of a commercial AI model by design. Diia.AI uses a commercial language model without ever exposing who the citizen is. Identifying details are removed and replaced with placeholders before a request reaches the model, then restored afterwards inside secure government systems. The model only ever handles anonymised requests, illustrating how a powerful external tool could be used without personal data leaving government control.

• The service is at an early stage, and scaling it brings challenges the team has flagged. For now, the assistant handles a limited set of requests, such as retrieving information on a citizen's business status, tax group, property, and pension and insurance details, and it is expected to do more as further government registries are connected. The European Commission's profile reports that the Ministry's AI Centre of Excellence has named the challenges wider adoption will bring: keeping performance and reliability steady as use grows, which will take sustained investment; the longer-term questions of cost, capacity and sovereignty that come with relying on an AI model run outside government; and the documentation and governance work still needed before the system can be reused elsewhere or brought fully into line with European regulations.

This case study was written with assistance from artificial intelligence.