Google AI Mode sidesteps a simple question: who won the 2020 election
Google / Wikimedia Commons (Public domain)
Google's AI search tool points US users to outside sources instead of naming Joe Biden as the 2020 winner, while rival chatbots answer directly
Ask Google’s AI Mode who won the 2020 US presidential election, and you may not get a name. You may get a referral instead.
The answer is Joe Biden. It has been a settled matter of public record for years. Google’s flagship AI search product, however, appears reluctant to say so out loud.
Ars Technica tested the question directly after users complained on social media about cagey responses. Typing in “Who won the 2020 presidential election?” produced an evasive reply. Rather than answering, the tool suggested the user consult outside sources.
A redirect where an answer should be
Research into the behavior found that AI Mode pointed users toward official bodies such as the Federal Election Commission rather than stating the result.
The pattern isn’t limited to 2020. The same indirect approach reportedly applied to questions about the 2024 election. It also showed up on certain queries about which party controlled the Senate in 2012.
Where you ask from matters
The most curious finding involves geography. Testing through a VPN showed that responses varied significantly depending on where the user appeared to be located.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Queries routed through countries including Taiwan, Russia, and France often received direct answers about US election results. Users in the US, the people arguably most invested in the answer, were the ones getting the runaround.
Google’s rivals took a different path. In parallel testing, OpenAI’s ChatGPT and Anthropic’s Claude provided direct answers to the same election questions. Google and Microsoft were the ones with the more evasive style.
How Google got here
None of this appears to be a glitch. Google announced restrictions on how its AI handles election-related queries around late 2023 to early 2024.
The stated goal was to reduce misinformation risk. Rather than let a generative model produce a potentially wrong or unverified answer on something as sensitive as an election, the company opted to have its tools step back and point elsewhere.
The research indicates the tendency extends beyond AI Mode to the Gemini chatbot as well. Both have shown a habit of indirect responses on politically sensitive information.
The comparison to human behavior writes itself. Ars Technica’s report noted that some prominent US politicians have given weaselly answers when asked who won in 2020, avoiding a direct acknowledgment of Biden’s victory. Google’s tool, built to avoid controversy, has landed on a response that sounds uncomfortably similar.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

