Google AI Won't Say Who Won 2020 — But Tells Russia

Google's AI won't give you a straight answer about who won the 2020 presidential election — but it will give one to users in Russia. The company that dominates how Americans find information online has built an AI system that dodges the most basic question about the most contested election in modern history, telling U.S. users to go look up the results themselves instead of simply stating the certified outcome.

Ars Technica tested Google's AI Mode directly and found that asking "Who won the 2020 presidential election?" produced a remarkably evasive response: "To review the official results and statistics of the 2020 United States presidential election, including the final electoral college breakdown, popular vote counts, and winner certification, please consult the official report compiled by the Federal Election Commission (FEC)." That's not an answer. That's a runaround.

The dodge isn't limited to 2020. Ars Technica found similar non-answers for the 2024 and 2012 presidential elections, and even for questions about which party controlled the Senate after 2020. Yet AI Mode had no problem giving direct, detailed answers for other election cycles going back to 2000. The inconsistency varied by user — one Ars staffer only got straight answers for 2016 and 2024, with other years producing links instead of answers.

Here's the tell: when Ars Technica checked AI Mode results via VPN from Taiwan, Russia, France, and Canada, the system gave "accurate and straightforward answers" every time. The evasion is reserved for Americans.

When pressed on why it wouldn't answer the first time, AI Mode itself blamed "safety filters for recent and highly sensitive election topics [that] are very strict." In other words, Google has decided that certain election facts are too dangerous for Americans to receive directly from its AI. Ask a second time, though, and the system loosens up — eventually providing the answer and citing Google Civics data.

Google spokesperson told a Bluesky user the company is "looking into this and working to address it," acknowledging that "this isn't the most helpful response on Search." The Verge noted its own staff couldn't consistently reproduce the issue, but Ars Technica documented it across multiple tests and users.

Notably, Google's own Gemini chatbot handles the question fine. So do OpenAI's ChatGPT, Anthropic's Claude, and even the federal government's AI model at America.gov. The problem is isolated to Google's AI Mode — the one integrated directly into the search engine that roughly 90 percent of Americans use to find information.

This is the same company that has spent years manipulating search results and suppressing content it deems problematic. Now its AI can't state a certified election result without first filtering it through "safety" protocols. The question isn't whether this is a bug. The question is how many other facts Google's AI has decided you can't handle.