Ars Technica observed Google AI Mode in the US directing questions about the 2020 presidential winner to FEC records rather than naming Joe Biden. Some other elections produced similar responses, but results varied by user, country and repeated questions; the standalone Gemini app answered directly. Google acknowledged a technical gap in verifying recent and historical election results and said it was addressing both 2020 queries and 2022 Senate queries. The model’s own explanation about safety filters was not a verified cause. The finding concerns particular Search responses, not all Google AI services.