The Chat GOV.BR is an interface tool that aims to provide and maintain a complete and integrated communication and relationship service with citizens regarding the provision of public services, encompassing multiple channels, digital service, automation, and the use of AI resources, including LLMs, proprietary to the generative IA platform or integrated from third parties, the construction of proactive communication journeys, and integration with government systems.
Initiative overview
The Chat Gov.BR was built as a secure platform to manage government AI chatbots using LLMs and RAG technology. This ensures millions of Brazilians get accurate and safe answers to public policies. It is essential because 144 agencies are preparing their own assistants. Our central infrastructure saves costs, protects data, and improves services.
In the pilot, the GOV.BR Chat helped high school students to enroll in graduation, clarifying doubts about the ENEM (the national exam, similar to "SATs") and SISU (the University selection system after ENEM results). The primary goal of this pilot project was to support high school students throughout their academic journey by providing clear guidance on the SISU system and other public policies. This initiative aimed to increase citizen confidence in official government communications while testing the ability of AI tools to handle massive peaks of demand when millions of people use the system simultaneously. Furthermore, the chatbot collected performance metrics and interaction data, which allowed leaders to make better, evidence-based decisions to continuously improve digital public services.
The initiative then evolved from a simple idea into a functional tool with a dedicated knowledge base. In January 2026, the government sent a proactive message to students to announce the ENEM results. This was a significant evolution because the chatbot moved from a passive tool to a proactive one. Instead of waiting for the student, the government started the conversation to clarify the next steps for SISU. This test showed that the government can share accurate information on a massive scale.
Looking ahead, this pilot is the foundation for a more modern and inclusive digital governance model. After the team assesses the final performance of this project, the goal is to institutionalize the use of AI in government communication. By using the lessons learned from this education case, the government aims to create "transactional chats." In the future, citizens will not only receive information but will also be able to request benefits and complete government tasks directly through the chatbot.
The initiative has yielded measurable results. In the Education case, we reached 6 million students, with a 3.5% interaction rate and an average of 5 messages per user. With Gov.br support, the chatbot independently resolved 65% of queries, significantly reducing human escalation. We measure these impacts using real-time platform analytics and resolution metrics.
Developing the GOV.BR chatbot brought several challenges. The main risk was the possibility of incorrect or confusing answers due to flaws in the knowledge base. To address this, we implemented a phased testing strategy: we started with internal checks and followed with a pilot group of students. These tests were essential to build the confidence needed for a large-scale launch.
Other relevant details
Key lessons include the need for better strategies to increase user engagement and the definition of clear success metrics. We also identified that analyzing millions of interactions requires an AI-powered analysis tool. For an initiative like this to succeed, a secure infrastructure and reliable data are vital. Looking ahead, we plan to scale the project into a permanent service and expand it to sectors like Health and Social Security. By applying these lessons, we aim to transform the pilot into a robust, government-wide AI ecosystem.
Looking ahead, we expect to transition from informational to transactional chats, allowing citizens to directly access policies and benefits. This will scale across many agencies, drastically reducing operational costs and improving service speed. These numeric outcomes prove the pilot's success in creating a more responsive and efficient digital government.
The Federal Government AI Platform is a strategic project to make AI more accessible, standardized, and faster to use across the government. Our long-term goal is to create a professional environment with advanced tools. We want to support all stages of an AI project, prioritizing open-source software. This allows us to work with partners while keeping our data safe in national infrastructure or approved clouds.
We expect many benefits from this platform, such as:
1. Delivering digital services to citizens faster;
2. Improving internal government work;
3. Saving money by reusing tools;
4. Promoting responsible innovation;
5. Training government public servants; and
6. Making better decisions based on data and ethics.
Right now, we are working on stages. Our first focus is Generative AI. We created a secure platform to manage 'virtual assistants' (chatbots) for the whole government. Our main goal is to build reliable and efficient assistants using advanced technology (like LLMs and RAG). This will help millions of Brazilians interact with the government and get accurate answers through different communication channels. This platform is very important because 144 government agencies plan to use chatbots soon (this is according to our annual monitoring with the federal government on the use of AI). Hiring companies in the Brazilian government is usually complicated. It was difficult to find a fast way to buy the services while also protecting our data and following all technical rules. To find the best solution, we studied international examples and did tests (Proofs of Concept) with big companies. We analyzed their tools based on our requirements for Generative AI. In the end, we decided to focus the project on a platform for messaging and chatbots to improve our relationship with citizens.
