Trump’s America.gov chatbot contradicted him on the 2020 election

The answers changed after the screenshot spread

The exchange circulated in a screenshot. After The Hill sought comment on Tuesday, an administration representative said the service would be supplemented with resources and “the most accurate information from the federal government.”

It was unclear what that meant. By the time we checked, America.gov would no longer answer questions about the 2020 election. Instead, it returned a standard response: “I help with U.S. government services and information about official programs. I do not discuss political issues or election results.”

Trump has continued to claim that his 2020 victory was stolen through Democratic fraud. The chatbot’s initial answer directly contradicted him. Trump recently directed that AI be called “superintelligence” in government documents and communications; that label does not make the system’s answers easier to control.

Elections were not the only point of friction

The New York Times reported that the chatbot also contradicted Trump on several other claims:

Responding to The New York Times, a White House representative offered the administration’s opposing position on the election, who ultimately bears the cost of tariffs, and the economic benefits of the “Working Families Tax Cuts” law.

A chatbot is not a message discipline tool

The administration has sought to reshape the federal government around Trump’s views, including by removing references to “woke” priorities from government websites and ending funding for climate research. Yet a chatbot drawing on government materials can still return answers that cut against those views. The administration’s response suggests one remedy is to add more resources and information, but it does not explain how that would prevent conflicting answers.

I think the more important question is what America.gov is meant to be: a guide to federal information, or a channel for the administration’s preferred account of events. Those are not the same job. Even without an obviously partisan agenda, an AI model can still make things up; with one, the challenge is not just accuracy, but deciding which answers the system is allowed to give.