Google's AI Mode avoids saying who won the 2020 US election
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
Ars Technica reported that, asked who won the 2020 presidential race, the AI Mode of Google's search engine gave no name, only a pointer to the official FEC report. The Gemini app did answer, and from other countries AI Mode replied plainly. Google says it is fixing the issue and blames a technical gap in how it handles 2020.
Google's AI Mode in Search, the answer generated with artificial intelligence inside the search engine, avoided naming the winner of the 2020 US presidential election, Ars Technica reported on October 28. The outlet ran the test after seeing complaints on social media about evasive answers: it asked in English who had won and, instead of answering, the tool told it to consult the official report from the Federal Election Commission (FEC), with the electoral college breakdown, the popular vote and the certification of the result.
Ars Technica got a similar response when asking about the 2024 election, whereas for other presidential elections going back to 2000 AI Mode did give detailed data. Other members of its staff found that direct answers and referrals to outside sources changed from one user to another. The Gemini app, also from Google, did say who had won, and so did ChatGPT and Claude. In spot checks over a VPN from Taiwan, Russia, France and Canada, AI Mode gave direct answers.
A Google spokesperson first said that the Gemini app answers that question more fully and that the search response was not the most helpful, and that the company was looking into it. The spokesperson later told Ars Technica that Google is fixing the problem with 2020 election queries in AI Mode. According to the spokesperson, its systems check election answers against news articles or data from the AP news agency for recent elections, and against historical records for older ones. The failure was put down to a technical gap with that year, which the spokesperson said also affects the 2022 Senate races and will be fixed in the same update.
Google has not published any official post on the matter: what is known of its position comes from the spokesperson's statements as reported by Ars Technica.
Why it matters · analysis and opinion
One manufacturer can give two different answers to the same question depending on the product, the user or the country the query comes from. Here the Gemini app answered while AI Mode pointed to an outside source, and Google has explained the failure as a gap in its verification system, not as a decision of principle. The lesson is practical: the safety layers placed on top of a model can leave the tool silent precisely on sensitive subjects. Anyone adding an AI assistant to a website, to customer service or to an internal process should test the sensitive questions of their sector on every channel and repeat the test after each update. What is still unknown is whether the announced fix reaches every user and every country.
Source: Ars Technica · Written with the help of AI: how we make the news
