Google's AI Mode provides indirect responses to queries about recent US presidential election winners, often directing users to the Federal Election Commission website. This behavior contrasts with the model's ability to provide direct answers for older elections dating back to 2000.
The inconsistency in responses varies by user and query. While some users receive direct answers for specific election years, others are redirected to external data sources. Similar patterns were observed when querying political history, such as Senate control following the 2020 elections.
The model's reluctance to provide direct answers appears to be location-dependent. Accessing Google's AI Mode via VPN from countries including Taiwan, Russia, France, and Canada yielded accurate and straightforward answers to questions about recent US election results.
When prompted repeatedly, the AI Mode eventually provides direct answers, citing Google Civics data. The system attributes the initial indirect responses to strict safety filters designed to prevent the spread of misinformation by directing users to official government databases.
Google stated that the Gemini app provides more comprehensive answers than the current Search AI Mode implementation. The company is investigating the issue to improve the quality of these responses.
Google's current public policy emphasizes surfacing authoritative election information and official resources. This follows a 2023 shift in YouTube's policy regarding content related to the 2020 election, aimed at balancing misinformation control with political speech.
