Google AI avoids naming winners of 2020 and 2024 elections

Evasive responses on recent polls

When users ask Google's AI Mode who won the 2020 presidential election, the system often avoids providing a direct name. Instead of identifying Joe Biden as the winner, the AI suggests that the user consult outside sources for the official results. This behavior has been noted by users on social media who noticed the AI's cagey approach to recent political history.

The AI's response for the 2020 election specifically directs people to the Federal Election Commission (FEC). It tells users to look there for the final electoral college breakdown, popular vote counts, and winner certification. This shift from a direct answer to a referral suggests a specific guardrail in place for recent American elections.

Similar behavior was observed when testing the AI with questions about the 2024 presidential election. Rather than stating a winner, the AI again pointed toward the FEC for the most accurate data. This pattern suggests that the AI is programmed to be particularly cautious regarding the most recent and ongoing political cycles in the United States.

Inconsistency across different years

The AI does not treat all historical elections with the same level of caution. Testing showed that the system was perfectly willing to provide detailed and correct responses for the winners of every other presidential election going back to the year 2000. This indicates that the AI's limitations are not a general inability to recall history, but are specific to certain timeframes.

For example, when asked about the 2012 election, the AI provided a link to the American Presidency Project’s 2012 Data Center. While this was not a direct statement of the winner, it differed from the specific FEC referral used for 2020. The AI seems to have a varying set of instructions for how to handle different decades of political data.

The behavior is not uniform across all users or even all staff members at the outlet reporting on the issue. Some testers reported getting immediate answers for the 2016 and 2024 elections, while others were met with links to outside data. This suggests that the AI's responses might vary based on individual user interactions or specific phrasing.

Broader political history filters

The evasiveness extends beyond just naming the winner of a presidential race. When asked which party controlled the Senate after the 2020 elections, the AI refused to give a straight answer. It instead told the user to check official congressional records and provided a link to the US Senate Party Division history page.

In contrast, the AI was much more direct when asked about the 1992 election. It stated clearly that the Democrats held a 57 to 43 majority in the Senate during that time. It even included Wikipedia as one of its references for this specific historical fact, showing no hesitation to provide a direct answer for older data.

These differences highlight a clear divide in how the AI processes information. It appears to be capable of providing specific numbers and party affiliations for the 1990s, but it defaults to bureaucratic referrals for the 2020s. This creates a tiered experience where the AI acts as a fact-provider for the past but a directory for the present.

How the AI manages sensitive data

The behavior described by Ars Technica suggests that Google may be using specific filters to manage political sensitivity. By directing users to the Federal Election Commission, the AI avoids taking a definitive stance on topics that have been subject to significant public debate. This is a common technique used to ensure the AI remains neutral on high-stakes issues.

The fact that the AI provides links to official government bodies like the FEC or the American Presidency Project shows a preference for authoritative sources. Instead of generating a summary, the AI acts as a bridge to primary documents. This allows the company to provide helpful information without being the primary narrator of the results.

For the user, this means the experience of using AI for research is becoming more nuanced. Depending on the year in question, the AI might act as a direct encyclopedia or a search assistant. This distinction is particularly noticeable when moving from 20th-century history to 21st-century current events.

What this means for AI users

For the average reader, these findings mean that AI models are not always neutral in their delivery of facts. While the information provided is generally accurate, the way it is presented can vary wildly based on the sensitivity of the topic. Users may find that the AI is more helpful for academic research into the past than for quick summaries of recent news.

This trend of