US Launches America.gov with Gemini and Grok for 29K Sites
The Trump administration unveiled America.gov, an AI portal responding to citizens using about 29,000 federal websites with Gemini and Grok. This design contrasts with the UK's GOV.UK Chat and Brazil's gov.br plans.

The Trump administration launched America.gov on Tuesday (29), a portal where citizens can ask questions in natural language and receive responses gathered from about 29,000 federal websites. The AI layer utilizes two competing models, Gemini from Google and Grok from Elon Musk's xAI.
Who Built It and Under What Rules
The project emerged from the National Design Studio, the White House's design and technology team led by Joe Gebbia, co-founder of Airbnb and the first Chief Design Officer for the U.S. government. In January, the same studio announced the goal of redesigning 27,000 federal websites and opened job listings for the task, according to Federal News Network and Architect Magazine. America.gov flips the order: instead of redesigning each website individually, it places a conversational intermediary over all of them.
The portal's privacy policy states that the AI providers operate under zero data retention agreements. Prompts and responses are not stored, and partners are contractually prohibited from selling information, creating advertising profiles, or training models with user data. Neither OpenAI nor Anthropic are among the suppliers mentioned at the launch.
From Search to Transaction
For now, America.gov informs and directs users. The government states that it intends to expand the tool so that users can start or complete tasks such as renewing a passport, enrolling in Medicare, checking student loan data, and booking federal camping areas. This second step increases the risk. A chatbot that delivers an incorrect answer generates a complaint; an agent who incorrectly processes Medicare enrollment leads to legal issues, and the announcement did not detail who would be accountable in such cases or whether there would be human review of transactions.
For integrators relying on federal contracts for website modernization, the architecture is more important than the chatbot itself. If 29,000 websites are now accessed through a single AI layer, the value shifts from each agency's front-end to the quality and structure of the content that feeds the model. Redesign contracts lose importance; content governance and legacy system integration contracts gain relevance.
Three Governments, Three Designs
The UK arrived earlier via a different path. The Government Digital Service officially launched GOV.UK Chat in the GOV.UK application in May 2026, following a restricted launch in March, according to the GDS blog and PublicTechnology. The British model is a single service, built by the government’s digital team, with responses limited to official content and personal data removed from queries. The American model outsources the engine to two private providers and aims for scale from day one.
In Brazil, gov.br already operates an AI chat service that routes specific questions to specialized agents in each area, and the federal government is discussing a shift from digital governance to an "agentic government," with an AI infrastructure platform for the Executive branch planned in the Brazilian AI Plan, according to a May report from Teletime. The difference in design reflects sovereignty: Brazil discusses its own infrastructure for models, while the U.S. opted to contract two commercial labs.
For CIOs and partners in consulting firms with public sector clients, comparing the three designs is more useful than debating model options. London gained control over content and risk by keeping the service in-house but at the cost of relying on an internal digital team that took over two years to launch. Washington traded that control for speed and competition between two suppliers, tying public service to contracts with laboratories that update model versions every few months. Brasília speaks of infrastructure autonomy, but the plan has yet to become operational at scale.
The Upcoming Test
The data the government did not disclose at the launch is the accuracy rate. The government cites the 29,000 sites as a convenience argument, but the number also highlights the problem: outdated pages, conflicting rules between agencies, and content never intended to be machine-readable. America.gov's performance will depend less on the choice between Gemini and Grok and more on the content cleanup that the National Design Studio announced in January, which has now shifted from a design project to a prerequisite for service.
Sources
- cnbc.comhttps://www.cnbc.com/2026/09/29/trump-ai-gemini-grok.html
- theregister.comhttps://www.theregister.com/public-sector/2026/09/29/trump-launches-americagov-with-ai-chatbots-at-its-core/5299907
- fedscoop.comhttps://fedscoop.com/trump-launches-ai-site-america-gov/
- america.govhttps://america.gov/privacy-policy
- federalnewsnetwork.comhttps://federalnewsnetwork.com/it-modernization/2026/01/national-design-studio-looks-to-overhaul-27000-federal-websites-and-is-hiring-a-team-to-do-it/
- gds.blog.gov.ukhttps://gds.blog.gov.uk/2026/05/14/gov-uk-chat-launches/
- teletime.com.brhttps://teletime.com.br/29/05/2026/governo-federal-avanca-no-uso-de-ia-e-planeja-plataforma-agentica/