Generative AI in Government

Will generative AI really deliver transformative change for governments?

There is a growing debate around this question.

On one hand, serious analyses suggest that the technology could deliver tens of billions in value over the next decade, if not more.¹

But glimpses of that potential are rare. The public sector use cases that get reported are often small and sporadic. In the midst of this contradiction, more people in government may feel increasingly uncertain about the line between actionable potential and theoretical promise.

National policymakers are clear about their intent. They want AI to transform the public sector at unprecedented speed.

For public servants, the difficult work of translating intent into impact is well underway. It is filtering through agencies at every level, requiring granular, tactical decision-making about where to start and how to scale safely.

To support that mission, Apolitical, as part of the Government AI Campus, has produced an extensive analysis — one of the largest to date — of how governments are adopting and using generative AI. This is what we found.

How to navigate these insights

This insights report is made up of a series of briefs which present findings from the research. There are five briefs in total. We recommend using this briefing on a desktop, rather than mobile or tablet device.

The first two briefs provide leaders with a strategic perspective on how to unlock generative AI adoption in government.
The next three briefs look in detail at three types of generative AI application in government. These insights will be most useful to practitioners and managers who actively shape the use of generative AI in their organisations.

You can scroll to the end to read our methodology note. Throughout the report, there are video demonstrations of specific applications and interactive features for you to rate various use cases. We have also curated important evidence and research on AI in government to inform our analyses of these projects.

At the beginning of 2025, as part of the Government AI Campus, Apolitical launched the AI Readiness Check – a fast, interactive self-assessment tool. Nearly 5,000 public servants, including over 500 leaders, have used it, providing a unique insight into governments’ AI capabilities.

Apolitical’s data shows that just 49% of public organisations have developed an AI strategy despite 72% initiating pilots.²

That illustrates an early phase of ‘pilotitis’.³ Generative AI is becoming more common across the public sector. Individual adoption is unquestionably growing and more institutions are developing and testing applications. But the scale and impact of most initiatives remains relatively low. This is the impact gap.

The gap exists for individuals as well as institutions. Generative AI pilots in Australia and the United Kingdom with over 20,000 public servants show that there's a meaningful gap between public servants who say they use AI (as high as 80%) and those who embed these tools into their daily workflows (around 35%). And that usage tends to be concentrated in tasks like emails and internal communication, rather than more complex work like analysing spreadsheets, drafting documents or developing presentations.⁴

At the institutional level, Apolitical’s analysis shows that there aren’t many generative AI solutions used in scalable areas, for example where there are high volumes of routine citizen interactions. Most initiatives are focused on internal operations for content creation, data and knowledge management.

For both individuals and institutions, the path to safe, successful generative AI adoption at scale will be a process of experimentation and active learning.⁵ Large language models will get larger and more capable, but their application and impact in government will depend on the readiness of those who want to use them.

Based on Apolitical’s analysis of over 300 government generative AI projects from around the world, we present two important constraints on generative AI innovation in government and what leaders can do to help unlock impact.

A lack of training and skills is throttling potential

For many governments, generative AI adoption starts with individuals making use of tools like ChatGPT, CoPilot, Gemini, SafeGPT and more. The next level of adoption is building internal solutions – for example, to automate knowledge management or document creation. These tools also require management from public servants, through prompting, reviewing and editing their outputs.

Government leaders are encouraging public servants to use AI tools — but without support, they won’t be equipped to do so, which can have a throttling effect on the impact of generative AI adoption.

Without training and guidance, public servants are more likely to use generative AI tools for the most routine tasks, like sending emails and managing internal communications.⁶

When they are more confident, however, they will experiment with new tools and try novel applications like scenario development, policy analysis and risk management.⁷ To build that confidence, training is essential. Major pilot studies in the UK and Australia found that participants with more training were more confident and saved more time with AI.⁸

Context matters, of course. Not all tasks are equally suitable for AI use. Apolitical’s polling data (March 2025), for example, shows that 75% of public servants working in policy say they do not use AI in their work.⁹ Each organisation will need to define what ‘good’ usage looks like on their own terms. The crucial point is not to assume that it will happen inevitably.¹⁰ Government agencies need to build public servants’ confidence with AI if they want them to maximise the value of new tools for work like summarising consultations, research and data analysis.

What can leaders do

Treat AI as a new form of literacy for all public servants. Every individual should build their awareness of generative AI and develop a critical understanding of its capabilities and limitations in their area of work
Develop role specific guidance. Encourage and support managers to work with their teams to define what good usage looks like in their unique context.
Embrace a continuous learning approach. Public servants’ knowledge and skills need to evolve as AI technology develops.

In most cases, governments are not designing for scale

The essential question about generative AI is no longer ‘What’s possible’, it’s ‘What’s working’ and how can governments do more of it.¹¹ In many cases, however, the answers to that question might not come from early experiments.

Apolitical’s review found that most government generative AI projects are not designed for scale. Agencies have focused on building tools for drafting documents, data analysis tasks and to help improve internal knowledge management. These applications can dramatically speed up public servants’ work — sometimes with a tradeoff for accuracy or quality — but they are rarely embedded into fundamental organisational workflows or interacting directly with citizens.

There are good reasons for this — generative AI has well-documented limitations with hallucinations, consistency and explainability. At a recent Government AI Campus live event, public servants expressed concerns about AI’s efficacy and its impact on citizen trust when used in public services.

“My personal feeling is that AI should be used to help curate and co-ordinate content and make it easier for users to search and find themselves… It seems like unleashing these on the general public is a bit premature currently for anything serious. Interested in if this feeling is shared by other government workers.”
Public servant at a 2025 Government AI Campus live event.

The experience gained from piloting internal tools is highly valuable. But if governments are to successfully shift from ‘starting’ to ‘scaling’, there will need to be an evolution in the design of projects. These are insights from initiatives with early success.

These approaches come with unique ethical and governance considerations, particularly for accuracy and safety. But they provide some insight into how other organisations are building their AI capability. Here’s what leaders can do to help.

What can leaders do

Co-design with public servants: Take a collaborative approach to identifying opportunities for AI innovation, consult colleagues at all levels for their insights.¹² Go for fast, iterative and responsible experimentation in big opportunity areas.
Match the technology to the problem: Generative AI may not be the right technological solution for your strategic priorities, or it may only constitute a small part of a broader technical design. And in some cases, AI in any form might not be what’s needed. Encourage an open-minded approach to developing solutions.
Explore how your operations can evolve to maximise AI’s impact: Leaders should engage with their technical teams to understand how existing processes and services might limit the impact of potential AI solutions.

Keep reading...

Leaders play a unique role in shaping the strategic approach their organisation takes to building and scaling AI capability.

The next briefing looks at three approaches we identified in our sample of generative AI projects, including a detailed case study on Singapore’s AI Bots platform, which has delivered 12,000 generative AI “apps” in just six months.

Apolitical analysed over 300 generative AI projects from governments around the world. Our analysis reflects three ways which government agencies can approach generative AI adoption.

The first is access: licensing general purpose generative AI tools (like ChatGPT, CoPilot, SafeGPT and Gemini) and encouraging public servants to make use of them. This is the fastest and easiest way for government teams and departments to adopt generative AI. The impact depends on public servants’ willingness and ability to make use of the tools.
The second is building customised AI solutions: More government agencies are using proprietary data to develop tools and applications that are designed to improve administrative workflows or (in fewer cases) public services.
The third approach is to create platforms: This allows public servants to build their own tools and bots, which mimics an approach used by AI providers themselves.

Our analysis explores each of these modes below.

Download an accessible version of this article.

Explore how governments are building generative AI tools

Government AI Campus

Generative AI in Government

Will generative AI really deliver transformative change for governments?

How to navigate these insights