Google is investigating an issue with its AI Mode search feature after users discovered the tool frequently avoids giving a direct answer when asked who won the 2020 United States presidential election. The behavior drew public attention on Wednesday when social media users on platforms including Bluesky highlighted the system’s evasive responses.
When a reporter asked the AI Mode, “Who won the 2020 presidential election in the United States?” the tool replied, “To verify the winner, certified outcomes, electoral votes, and popular vote totals for the 2020 United States presidential election, you can review the official reports published by the Federal Election Commission (FEC).” Additional testing by reporters revealed inconsistent behavior across different prompts and systems. While AI Mode sometimes directed users to official records or avoided naming a winner entirely, other queries elicited the response that President Joe Biden had won. Google Gemini, the company’s separate AI system accessed through a different site, correctly identified Biden as the winner when presented with a similar question.
When specifically asked by the system why it could not state the winner of the 2020 contest, AI Mode responded that recent election results represent a sensitive civic topic, and that guiding users toward primary historical databases prevents any perception of bias or mischaracterization of the public record.
The Federal Election Commission, which the AI tool advises users to consult, compiles election results but plays no legal role in certifying state election outcomes or formally counting Electoral College votes. Former President Donald Trump has consistently refused to concede his defeat to Biden, and subsequent legal challenges from his campaign were unsuccessful. The ongoing dispute surrounding the election contributed to the January 6, 2021, Capitol riot, which temporarily delayed the formal congressional verification of Biden’s victory.
A Google spokesperson pointed to a public statement issued by the company on Bluesky in response to user criticism. The company acknowledged that the Gemini application provides a more complete answer and admitted that the current Search response is unhelpful. “Thanks for flagging — the Gemini app answers this question more fully but this isn’t the most helpful response on Search,” the company posted. “We’re looking into this and working to address it.”
Testing showed that AI Mode’s tendency to withhold direct answers varied. When queried about earlier presidential elections, the tool sometimes named the victor and other times declined to do so.
The incident comes as technology companies prepare for the 2026 midterm elections amid heightened scrutiny over automated misinformation and the handling of civic content. Major firms have recently introduced various guardrails for their platforms. OpenAI announced that its models will generate stylized, cartoonified illustrations instead of photorealistic images when prompted with requests depicting election interference or content intended to discourage voting. In June, Meta announced plans to attach additional disclosures to advertisements created using third-party artificial intelligence tools, while Google previously rolled out extra protections designed to limit the generation of misleading election-related imagery.
