Originally published on The Daily Flare.
Google AI Mode directed a user asking who won the 2020 U.S. presidential election to Federal Election Commission materials rather than naming Joe Biden, according to Ars Technica.
An indirect answer to a direct query
Ars Technica reported that its reporter entered the question, “Who won the 2020 presidential election?” into Google AI Mode on October 28. The system replied by directing the user to the FEC’s official report for election results, vote totals, the Electoral College breakdown and winner certification.
The response did not provide the requested winner in its own answer. Instead, it pointed toward an external source for the information. Ars Technica said the result was unexpectedly indirect, particularly because the feature is designed to respond to user queries with synthesized information.
The same reporter received a similarly indirect response when asking about the 2024 presidential election. A question about the 2012 contest prompted a link to the American Presidency Project’s data center. By contrast, the system gave detailed answers about other presidential elections dating back to 2000.
Those results were not consistent across all users. Ars Technica staff members found that which election questions received direct answers appeared to vary between accounts. One staffer, for example, received immediate answers for the 2016 and 2024 elections, while other questions about contests from 2000 onward led to outside sources instead.
Responses changed by question and location
The uneven pattern also appeared in a question about control of the U.S. Senate after the 2020 elections. In that test, Google AI Mode referred the user to official congressional records and the Senate’s party-divisions history. Asked about the 1992 election, it gave a direct answer, saying Democrats held a 57-to-43 majority, and included Wikipedia among its references.
Ars Technica reported that other political-history questions produced different styles of response, with some receiving direct answers and others links to reference material. The results suggest that the feature did not handle every election query in the same way during the outlet’s testing. They do not establish that every user, or every version of the system, would receive the same replies.
The report also found that Google AI Mode answered more directly in spot checks made through VPN connections in Taiwan, Russia, France and Canada. In those tests, it gave accurate, straightforward responses to multiple questions about recent U.S. election results. Ars Technica did not describe those checks as a comprehensive test of the service in each country.
Google’s separate Gemini model answered questions about recent presidential winners directly in the reporter’s tests. Ars Technica said OpenAI’s ChatGPT, Anthropic’s Claude and the government-run AI model at America.gov also supplied direct responses. Those comparisons involved the reporter’s queries and do not establish how the systems would respond in every circumstance.
Repeated prompts produced different results
The reporter found that asking the same questions again could change Google AI Mode’s response. After a repeat prompt, the system named winners for the elections it had previously handled with links, citing Google Civics data. This behavior added another variable to the test: users could see a referral first and a direct answer after asking again.
The account leaves open how consistently Google AI Mode applies its approach to election questions. The reported tests showed variation by election year, user and location, while repeated prompts sometimes changed the result. For readers seeking an authoritative record, the system’s own first response directed them to the Federal Election Commission’s official election materials.
Top comments (0)