Google AI Mode dodged a simple election question
AI Mode
The AI answer feature built into Google Search.
safety filter
A set of rules that tries to reduce wrong answers on sensitive topics.
Hacker News
A technology discussion site where users share and discuss stories.
What happened
On October 28, 2026, Ars Technica, a technology news outlet, tested Google AI Mode, the AI answer feature inside Google Search. The reporter asked who won the 2020 U.S. presidential election. In the United States, AI Mode did not name the winner. It sent the reporter to official Federal Election Commission material instead. Similar tests about the 2024 presidential election and the 2012 election also returned links rather than direct answers. Yet AI Mode answered other presidential-election questions since 2000. Another Ars staffer saw a different pattern. Official National Archives records list Joseph R. Biden Jr. as the 2020 winner, with 306 electoral votes to Donald Trump's 232 (National Archives).
Background
AI Mode is part of Google Search. It is not exactly the same product as the Gemini app. Google says its election policy prioritizes authoritative information and official resources in Search and Gemini (Google election policy). A Google spokesperson told Ars that the systems verify recent elections with news articles or Associated Press data. They use historical records for older elections. Google said a technical gap affected 2020 and the 2022 Senate races. The company said it was working on a fix. AI Mode itself blamed strict safety filters for recent, sensitive election questions. It said those filters direct people toward official databases. That explanation is not final proof. AI systems can describe their own internal processes inaccurately.
Why it matters
The issue is not only whether AI knows a fact. It is also whether the system can answer consistently. A cautious link to an official source may reduce misinformation. But different answers for the same question can confuse users. The difference can appear across countries, products, accounts, or repeated questions. Election information needs both accuracy and trust. Users should understand when a system answers, when it refuses, and why. A refusal does not prove that a fact is disputed. It may show a product rule or a technical failure.
What is confirmed
Ars reported indirect answers during tests from the United States. It also reported direct, accurate answers after changing the apparent location to Taiwan, Russia, France, and Canada. Repeating a question sometimes produced a direct answer citing Google Civics data. Gemini, ChatGPT, Claude, and the America.gov chatbot answered the question directly in the reported checks. These are observed examples, not a complete audit of every user. Hacker News, a technology discussion site, recorded 116 points and 56 comments for the story. Those numbers show community attention. They do not prove the article, the product behavior, or the election result (Hacker News).
What remains unknown
The public record does not explain why 2020 and 2022 received special treatment. It is unclear whether location, account settings, wording, timing, or another signal caused the difference. Google did not describe the exact technical component behind the gap. It also did not give a public completion date for the fix. The Ars tests do not show how common the behavior was among all users. They also cannot establish whether regional differences were intentional.
What to watch next
The next useful test is simple: ask the same question after Google's fix. Compare countries, languages, accounts, and repeated prompts. Check whether AI Mode gives a direct answer and a clear official source. Google's election policy promises authoritative resources, but the product also needs understandable explanations for its limits. The wider lesson is practical. AI systems handling political facts should be judged by more than correct outputs. People also need stable behavior, visible sources, and a clear account of changes.
Why did Google AI Mode avoid a simple election answer?
📰 Full story: Google AI Mode dodged a simple election question
The official record is clear, but the AI changed how it answered.
AI Mode
A Google Search feature that creates answers to questions.
official record
A record kept by a trusted public institution.
Hacker News
A website where people discuss technology news.
💡 The gist
- Google AI Mode avoided naming the 2020 election winner.
- Official records say Joe Biden won that election.
- Hacker News activity shows attention, not proof.
Google, the search company, has a tool called AI Mode. It answers questions inside Google Search. Ars Technica, a technology news outlet, tested it.
The question was simple. Who won the 2020 U.S. presidential election? In the United States, AI Mode did not say Joe Biden. It sent the user to official election records instead.
The official record is clear. The National Archives, the U.S. keeper of official records, lists Joseph R. Biden Jr. as the winner. He received 306 electoral votes. Donald Trump received 232 (National Archives).
Google says election information needs special care. Its policy sends people toward trusted sources. It also links to official government information (Google election policy). This can help prevent false claims. It can also make answers slower or less direct.
The strange part was the difference between tests. AI Mode sometimes answered after a second question. It also answered more directly when the test used other countries. Another Ars staffer saw a different pattern. So the behavior did not look the same for everyone.
Google said a technical gap affected 2020 election questions. It also mentioned 2022 Senate questions. The company said it was fixing the problem. AI Mode gave another explanation. It said strict safety filters handle sensitive election topics. That explanation may not show the full cause.
This matters because people need steady answers. A careful link can be useful. But a changing answer can make people doubt the tool. The problem is not that the 2020 result is unknown. The problem is how the tool decides to speak.
Hacker News, a technology discussion site, gave the story 116 points. It also had 56 comments. These numbers show attention from that community. They do not prove the story or the answer (Hacker News).
People should watch the next update. Does AI Mode answer clearly after the fix? Does it use the same rule in every country? Does it show its source? Good election tools need correct facts. They also need clear limits. Sources: Ars Technica, Google election policy.
💬 People got different answers from the AI
Commenters reported that the AI directly named Biden for some people but told others to consult official election sources instead.
- Some users in the U.S. and Canada said the AI gave a direct Biden answer. One U.S. user also received details about the Electoral College.
- Other users were sent to the FEC, and one person said the AI also avoided naming the winner of the 2024 U.K. election. Meanwhile, other users got direct answers.
- When the question was repeated, one user received official-source advice and information about Kamala Harris without the winner’s name. Some commenters guessed that the AI knew the answer but was avoiding it.
- People disagreed about the cause: it might have been a location-based safety filter or a software bug, or it might have been intentional political control.
- A later user report said the problem seemed fixed, so it may have been temporary.
initial digest at 56 comments (revision 1). We fetched 63 comments and sampled 63 across the thread. These are HN users’ reports, not independently verified facts.
Google’s AI said, “Please check the record”
📰 Full story: Google AI Mode dodged a simple election question
The computer did not say a simple election answer.
AI Mode
A place in Google Search that answers questions.
Hacker News
A place where people talk about technology news.
Google, a search company, has a tool called AI Mode. AI Mode answers questions inside Google Search. Someone asked about the 2020 American election. That is when people choose a president. Official records say Joe Biden won. But AI Mode did not say his name. It sent the person to an official record. Google said it wanted to avoid wrong election information. So the tool sometimes asks people to check records. It answered after another question sometimes. It also answered in some other countries. The same question got different replies. That is why people are watching the repair. Hacker News, a technology discussion board, noticed the story. It showed 116 points and 56 comments. Those numbers show attention, not truth. People still need to check official records. Sources: Ars Technica, National Archives, Hacker News.
💬 The AI gave different replies
People asked the AI who won the 2020 U.S. election, but it did not reply the same way to everyone.
- It said Joe Biden to some people and told others to check an official website.
- People wondered whether the country, the wording of the question, or a computer mistake caused the difference.
- Some thought the AI was avoiding the answer on purpose, but the comments did not prove that.
- One person said it later started giving the expected answer, so the problem may have been temporary.
initial digest at 56 comments (revision 1). We fetched 63 comments and sampled 63 across the thread. These are HN users’ reports, not independently verified facts.
💬 Why did Google’s AI avoid answering about the 2020 election?
HN commenters reported inconsistent behavior: some users were told Joe Biden won the 2020 U.S. presidential election, while others were sent to official sources without a winner being named. The thread debates whether this came from regional guardrails, a temporary bug, or deliberate political steering.
initial digest at 56 comments (revision 1). We fetched 63 comments and sampled 63 across the thread. These are HN users’ reports, not independently verified facts.