Google's AI Mode often won't name the 2020 US election winner for US users, but answers abroad

Days before the US midterm elections on November 3, several outlets have found that AI Mode, the conversational search feature in Google Search, frequently declines to say who won the 2020 presidential election when asked from the United States. The behaviour was documented on October 28 by Ars Technica, Fast Company and TheWrap. Google's separate Gemini app, by contrast, named Joe Biden.

What the tests found

Ars Technica's Kyle Orland reported that AI Mode referred his questions about the 2020 and 2024 winners to Federal Election Commission material, and sent a 2012 question to an academic archive, while answering other presidential races back to 2000. A second Ars staffer got direct answers only for 2016 and 2024. When Orland repeated the questions, AI Mode named the winners, citing Google Civics data, so the results were inconsistent rather than a fixed rule. Spot checks through VPNs in Taiwan, Russia, France and Canada returned direct answers.

Fast Company saw a similar pattern. In the US, AI Mode deflected both "who won" and "was it stolen" questions about 2020, pointing to the FEC, while stating plainly that Donald Trump won in 2016 and 2024. One US-based staffer needed eight attempts to get an answer naming Biden. The article's UK-based author received a direct answer, then got the same deflection after switching to a US VPN connection. Ars reported that Gemini, ChatGPT and Claude answered the 2020 question directly.

Google's response

A Google spokesperson pointed reporters to a reply the company posted on Bluesky. It says the Gemini app answers the question more fully but that the Search result "isn't the most helpful response," and that Google is looking into the issue and working to address it. Google has not explained why answers differ by country or why AI Mode and Gemini behave differently. AI Weekly suggests the hedging may stem from restrictions on election queries that Google introduced in late 2023 and early 2024; Google has not confirmed that.

Why it matters

This is not new territory: in 2024, tests reported by the Washington Post found Gemini and Microsoft Copilot refusing similar questions, with both companies saying the caution was deliberate. The timing is sensitive now. TheWrap notes that Google said in September it was rolling out election information tools across Search, AI Mode and Gemini, and Fast Company points to renewed voter-fraud claims ahead of the vote. An AI system that hedges on a settled result can leave room for doubt, even when the reason is overcautious design rather than intent.