Google's AI Mode is refusing to say who won the 2020 US presidential election. When Ars Technica asked the search feature the direct question, it did not name Joe Biden. Instead, it advised consulting the official report from the Federal Election Commission. The same pattern appeared for the 2024 election and, in some cases, for 2012.
The odd behavior drew attention after users complained on social media. Ars found that the refusal is inconsistent: AI Mode answered other election questions normally, and different users got different results. Google says it is now fixing the issue with 2020 election queries in AI Mode.
What Google AI Mode told users
In Ars Technica's tests, AI Mode answered questions about most presidential elections going back to 2000 with correct, detailed responses. But for 2020, it pointed to the FEC rather than stating the result. For 2012, it linked to The American Presidency Project's data center. Other Ars staffers saw different combinations of answered and deflected questions.
The pattern extended beyond presidential results. Asked which party controlled the Senate after the 2020 elections, AI Mode said to check official congressional records. Asked the same question for 1992, it answered directly, noting a 57 to 43 Democratic majority and citing Wikipedia among its sources.
Google's own Gemini model answered the election questions directly, as did ChatGPT, Claude and the government-run AI model at America.gov. So did AI Mode itself when Ars tested it via VPN from Taiwan, Russia, France and Canada.
Asked twice, AI Mode became more forthcoming. It gave the winner for the three elections it had previously dodged, citing Google Civics data. It blamed strict safety filters for recent and highly sensitive election topics, and said the system is automated to send users to official, certified government databases to avoid unverified data or misinformation.
