U.S. President Donald Trump announced the launch of America.gov, a conversational interface intended to simplify access to government information and services, with Google and the Gemini model participating, and Grok also being used, according to a government official. The project raises questions about the risks of hallucinations when users rely on model answers for visas, taxes, and benefits.
U.S. President Donald Trump announced on September 29, 2026, the launch of America.gov, an artificial intelligence-powered conversational service to help citizens find government services and related information. The initiative presents itself as a single point of entry, rather than requiring users to navigate tens of thousands of government websites, rules, and procedures.
Google confirmed its participation in the launch and said that its Gemini model is used within the service. Joe Gebbia, the U.S. chief design officer, added that the government also used Grok. The source did not clarify the scope of the technical integration between these models or the mechanisms America.gov will use to verify answers before displaying them to users.
Why does this news matter?
The initiative targets a real problem: the difficulty of finding correct government information amid scattered websites and procedures. But turning this task over to an interface that relies on large language models shifts part of the problem to a different question: how accurate the answer received by the user is, and how clear its source and currency are.
The article cautions that language models are not infallible and may produce fabricated or incorrect information. In government services, an error may not remain merely an annoying answer; a user who relies on inaccurate information about applying for food assistance, renewing a visa, or filing a tax return could miss a deadline, be denied a benefit, or face a penalty.
What changes in practice?
If America.gov succeeds in directing users to official information in an understandable way, it could reduce the time needed to find the appropriate forms and steps. But the source provides no details about human review procedures, how the original government links will be displayed, or how cases that vary according to the state or the user's legal status will be handled.
The article points to an example reported by CNN involving the U.S. military mistakenly believing that a Chinese vessel was carrying components for nuclear weapons, before an armed operation was canceled at the last minute after it was discovered that the presumed threat had resulted from an artificial intelligence hallucination. This does not mean that America.gov will lead to a similar outcome, but it illustrates that errors in systems based on models can extend to decisions with significant consequences.
Open questions
The announcement alone does not establish that America.gov can provide reliable guidance in every case. The service's quality will remain tied to its ability to update information, attribute answers to official sources, alert users when an answer is uncertain, and provide a clear path for direct verification before taking legal or financial action.