Why Wouldn’t Google Tell Me Who Won the 2020 Election?

The question "Who won the 2020 presidential election?" should have been an easy one for Google's AI Mode. However, when the reporter inquired about the outcome earlier, the AI tool remained silent. The response only directed users to the Federal Election Commission's certified reports, without providing any links, citations, or additional information.

This puzzling behavior was observed across various social media platforms, with users sharing screenshots of AI refusing to declare Joe Biden as the 2020 victor, even as it readily identified Donald Trump as the winner of the 2016 and 2024 elections. The meme-worthy incident prompted users to come up with creative ways to trick the AI into revealing the truth.

For instance, asking the AI to spell the winner's name backward led to the correct answer: "nebiB eoJ won the 2020 United States presidential election." When confronted with "what the hell," the AI conceded the correct result and apologized for its initial reluctance to provide the information. While AI tools giving strange answers is not new, this particular instance of an influential AI platform failing to provide a straightforward factual question is unusual.

Many other AI models, including ChatGPT, Claude, Microsoft's Copilot, X's Grok, and even the Trump administration's America.gov chatbot, correctly stated that Biden won the 2020 election. The issue extended beyond Google's AI Mode, as a standard Google search for the same question also failed to display an AI summary at the top of the page, only presenting links to other websites.

Notably, a standard query about the 2016 presidential election did trigger an AI answer, which raised suspicions among users. Trump continues to claim that he won the 2020 election, and questions about the result have become a litmus test of loyalty to the president. Google has a history of altering its products to align with Trump's preferences, such as renaming the Gulf of Mexico as the "Gulf of America" and Lake Ontario as "Lake America."

At a recent rally, Trump reportedly called Google CEO Sundar Pichai to inquire about renaming the Atlantic Ocean as well. However, there is no evidence suggesting Google intentionally censored its chatbot. When the reporter inquired about the bizarre result this morning, Google stated that it was investigating the issue. By afternoon, Google revealed that it had identified the problem and was working to fix it.

The company noted that its AI search tool verified election responses using news articles or AP data for recent elections, while historic records were used for older ones. A spokesperson for Google declined to provide further details. Despite Google's assurance that this was unintentional, the incident highlights the opaque and unpredictable nature of large language models.

The company's AI chatbots have been trained to be cautious around sensitive topics, including election results, which can sometimes lead to unexpected outcomes. While Google asserts that its AI overviews reach over 2 billion people monthly, the fact that the bots struggled with an elementary election question raises concerns about the potential for subtle errors in other areas.