Google’s AI Mode sometimes declines to state who won recent US presidential elections and instead points users to official records, according to testing by Ars Technica. A direct question about the 2020 result produced a link to the Federal Election Commission rather than naming Joe Biden.
The behavior was inconsistent. Other Ars staff members received direct answers for some election years and evasive links for others, with the pattern varying by user. Questions about older contests were more likely to produce a straightforward winner. Similar caution appeared in a question about which party controlled the Senate after the 2020 elections, while a 1992 version received a direct answer.
The responses were not factually false, and linking to authoritative records can be appropriate for disputed or high-stakes information. The problem is that selective refusal can make a basic historical fact look uncertain, especially when the same system answers comparable questions from other periods without hesitation.
The testing does not reveal whether the behavior comes from policy rules, retrieval results, personalization or an experiment. Google would need to explain the criteria and make them consistent. For users, the episode is another reminder that an AI search interface may silently change from answering a question to deciding which questions it prefers not to answer.