Google AI Mode Hesitates to Confirm 2020 US Election Winner Amid Safety Measures

Image: Ars Technica · Source

Google's AI Mode avoids directly identifying the 2020 US presidential election winner, instead directing users to official sources—a caution not seen in other AI models or for elections outside the US, reflecting strict safety filters on recent sensitive topics.

Google's AI Mode exhibits notable caution when asked about the winner of the 2020 US presidential election, giving indirect responses that direct users to official resources rather than naming Joe Biden explicitly. Ars Technica reported that upon asking "Who won the 2020 presidential election?" Google AI Mode replied by suggesting consulting the Federal Election Commission's official report instead of providing a direct answer. This pattern extends to other recent elections, including 2024 and certain Senate races, where the AI prefers linking to official data rather than stating winners outright.

Interestingly, this reluctance appears specific to certain recent US elections and varies among users. Some Ars staffers observed that Google AI Mode provided direct answers for other presidential contests as far back as 2000 but remained evasive for 2020 and some recent cycles. However, when queries were made from outside the US via VPNs—such as from Taiwan, Russia, France, and Canada—the AI offered straightforward and accurate responses about these elections, which indicates regional or contextual filtering.

Other AI models, including Google's Gemini, OpenAI's ChatGPT, Anthropic’s Claude, and a government-run model at America.gov, readily provided direct and detailed answers about recent US election outcomes. The difference with Google AI Mode seems tied to its "safety filters for recent and highly sensitive election topics," which Google says are "very strict" to prevent misinformation. According to a company spokesperson, these filters automate directing users to certified government databases for recent elections instead of displaying potentially unverified information.

Google emphasizes its commitment to providing authoritative election data through official channels. In preparation for the 2026 US midterm elections, Google is enhancing tools in Search and AI products like Gemini app and AI Mode to highlight official polling locations, registration deadlines, and real-time results from trusted organizations such as The Associated Press. Google further applies technologies to detect and label AI-generated content to maintain transparency and reduce misinformation risks.

The cautious approach by Google AI Mode may also be influenced by the company’s broader policies against election misinformation, including previous YouTube decisions to moderate content making false election claims to balance free speech and safety. A Google spokesperson acknowledged a technical gap affecting responses about the 2020 election and some 2022 Senate races, which the company is actively fixing.

In summary, Google AI Mode's sidestepping of directly naming the 2020 election winner reflects Google’s stringent automated safety protocols aimed at steering users to official government data for recent elections to minimize misinformation risks. This results in behavior distinct from other AI systems and varies based on user location and query repetition, illustrating the complexity of moderating sensitive political information through AI.