Google's AI Mode Won't Say Who Won the 2020 Election, Calls It a Technical Gap

Ask Google's AI Mode who won the 2020 presidential election and, for many users, the answer is a shrug in text form. Rather than naming a winner, the AI-powered search experience has been directing people to consult the Federal Election Commission's official report. The odd behavior, first detailed by Ars Technica, has drawn enough attention that Google now acknowledges it as a technical gap it is working to fix, less than two weeks before the US midterm elections.

The episode is a useful reminder that AI search products can fail in strange ways even on settled historical facts, and that staying current on AI developments increasingly means tracking not just model launches but how these systems behave under political pressure.

What Ars Technica's Testing Found

When an Ars Technica reporter put the direct question to Google's AI Mode, the response was strikingly indirect: to review the official results and statistics of the 2020 United States presidential election, including the final electoral college breakdown, popular vote counts, and winner certification, the tool recommended consulting the official report compiled by the FEC.

The dodge was not consistent. Testing showed similar indirect answers for the 2024 and 2012 elections, while AI Mode gave detailed, correct answers for every other presidential election going back to 2000. Different Ars staffers saw different patterns, with one getting immediate answers only for the 2016 and 2024 races and links to outside data for everything else.

Political history questions beyond presidential races were affected too. Asked which party controlled the Senate after the 2020 elections, AI Mode told the reporter to check official congressional records and linked to the Senate's party division history page. Asked the same question about the 1992 election, it answered directly that Democrats held a 57-to-43 majority.

Other Chatbots Answer Without Hesitation

Google's own Gemini model, separate from AI Mode, answered the question without issue, according to the report. So did OpenAI's ChatGPT, Anthropic's Claude, and even the government-run AI model on America.gov. That distinction matters, because it suggests the problem lies in how AI Mode's search-layer safeguards handle election queries rather than in the underlying language model's knowledge.

The behavior also changed with geography. Ars Technica's spot checks through VPNs in Taiwan, Russia, France, and Canada all returned accurate, straightforward answers about recent US election results, implying the restrictive handling is specific to users in the United States.

Google's Response: A Technical Gap

A Google spokesperson first pointed to a social media post acknowledging that the Gemini app answers the question more fully and that AI Mode's response was not the most helpful, adding that the company was looking into it. The spokesperson later gave a more concrete explanation, telling Ars Technica that Google is fixing the issue with 2020 election queries in AI Mode. According to the spokesperson, Google's systems verify election responses using news articles or AP data for recent elections and historic records for older ones, and there was a technical gap with 2020, with the same true for 2022 Senate races, which the company is working to fix in the same update.

An interesting wrinkle from the testing: persistence paid off. When the same questions were asked a second time, AI Mode provided the winner directly, citing Google Civics data. Asked why it had not answered initially, the tool blamed strict safety filters for recent and highly sensitive election topics and described a heavily automated system that routes such queries to official certified government databases to prevent any risk of displaying unverified data or misinformation. As Ars Technica notes, AI models are unreliable narrators of their own internal processes, so that explanation deserves some caution.

Why Election Answers Are Hard for AI Search

Google is not wrong to treat election queries as high-stakes. Generative search systems can hallucinate, and election misinformation is a documented problem that platforms have repeatedly stumbled on. In 2023, Google-owned YouTube reversed a policy that barred all videos making false claims about the 2020 election, saying removals could curtail political speech without meaningfully reducing real-world harm.

The company's current public posture emphasizes surfacing authoritative election information directly within results, including direct links to official resources, according to its Election Public Policy page and a September blog post about helping voters find reliable information around the midterms. The AI Mode incident shows the gap between that policy and the on-the-ground behavior of an automated system: in trying to avoid saying something wrong, it ended up refusing to say something settled.

With the midterms imminent, the episode illustrates the fine line AI products walk during elections. Over-blocking factual answers frustrates users and can erode trust just as surely as errors do, and a dodge that varies by user, country, and repetition is hard to defend as either safety or accuracy. Google's promised update will show whether the fix restores direct answers uniformly, or whether election queries in AI Mode remain a lottery of filters and technical gaps.

Stay Ahead of AI

For continuous coverage of AI policy, platform behavior, and the systems shaping online information, follow AI Buzz Wire for breaking AI news and independent analysis.

Read more AI news at AI Buzz Wire