Google Search's AI Won't Name the 2020 US Election Winner. Google Says It's Working on a Fix
Users in the US were told to check Federal Election Commission records instead of getting the winner's name. Google admits there was a gap in how it verifies such answers.
There have been cases where Google Search’s AI Mode did not directly name the winner when asked who won the 2020 US presidential election. The US tech outlet Ars Technica reported the problem on October 28 local time. Google acknowledged that the response was unhelpful and said it is working on a fix.
What happens when you ask who won
When accessed from the United States, AI Mode pointed users to records from the US Federal Election Commission (FEC) instead of naming Joe Biden. Even though the question concerned a result that is part of the official record, users were sent off to look up other sources. According to the US National Archives and Records Administration’s tally of the 2020 Electoral College vote, Biden received 306 votes and Donald Trump 232. Biden won.
The answers varied from person to person and from attempt to attempt. Ars Technica’s reporters each ran into a different election year that the tool would not answer about, and in some cases asking the same question again produced the winner’s name. When the question was asked from outside the United States, Ars Technica reported, the AI answered more directly.
Even within Google’s own products, the answers diverged. Unlike Search’s AI Mode, the Gemini app named the 2020 presidential election winner right away. ChatGPT and Claude also named the winner. Google itself said that users can get a more complete answer to the same question in the Gemini app.
How Google explained the cause
Google said there was a technical gap in the process it uses to verify election-related answers. The company explained that it checks the accuracy of answers about recent elections against news articles or AP data, and answers about older elections against historical records. Something went wrong in that process when handling the 2020 presidential election and the 2022 Senate elections, Google said. The company added that it is fixing the issue.
Source: Google (press release)
Ahead of the 2024 elections, Google had previously announced that it would limit the range of election questions answered by Bard, as it was then called, and by generative AI in Search. This time the company attributed the problem not to those limits but to a technical issue in the verification process for specific elections.
Would the same thing happen in Korea?
AI Mode is a search feature available to users in Korea as well. Google began rolling it out in Korea in phases from September 2025, with Korean-language support. That said, the cases covered in this report were mostly from US-based access. Determining the impact on Korean users would require checking whether the same answers appear when accessing the service from within Korea and when asking in Korean.
The episode shows that even for questions with a clear official record, AI search can hand users other sources instead of an answer, and can respond differently to the same question. It is a practical problem, because someone trying to confirm an election result has to run another search. What to watch next is whether, once Google completes its fix, the tool consistently names the winner of the elections in question, and whether the differences based on where users are connecting from disappear as well.