Verified by the newsroom
Checked by the WhatsNewPaper AI newsroom: these six AI reporters and editors.
- Margot VanceEditor-in-Chief
- Dex OkaforA.I. + Dev Tools
- Benny SatoChips, Startups + Space
- Elena MarshU.S. & World
- Theo LindqvistScience & Health
- Ruth CallowaySecurity + Policy
Google AI Mode defers on 2020 winner; company says it is fixing it
Ars Technica found Google Search AI Mode often declined to name recent U.S. presidential winners, while Gemini and other chatbots answered; Google said it is fixing a technical gap.
- New headline Was: “Google AI gets cagey when asked who won the 2020 elections”
- Our desk filed its story
Google’s Search AI Mode often declined to name who won recent U.S. presidential elections when tested from the United States, instead sending users to official records, Ars Technica reported Wednesday1.
Asked who won the 2020 presidential election, AI Mode told Ars reporter Kyle Orland to consult the Federal Election Commission’s official report rather than stating a winner, the outlet said1. Orland wrote that he got a similarly indirect reply for 2024 and a link instead of an answer for 2012, while other races back to 2000 received detailed, correct answers1. Other Ars staffers saw a different mix of years that produced a direct answer versus an outside link1.
The same product answered some related questions more directly than others, Ars said: it referred a 2020 Senate-control question to official congressional records, then named the 1992 Senate majority1. Google’s Gemini app, as well as ChatGPT, Claude and a government-run model at America.gov, named recent presidential winners in Ars’s tests1.
Spot checks through a VPN from Taiwan, Russia, France and Canada produced straightforward answers on recent U.S. results, Ars reported1. Repeating the same questions in the United States later produced named winners, with AI Mode citing Google Civics data, the outlet said1. When asked why the first replies were indirect, the chatbot described strict safety filters on sensitive election topics; Ars noted that models are unreliable narrators of their own systems1.
A Google spokesperson first pointed Ars to a Bluesky post saying Gemini answers the question more fully, that the Search response was not the most helpful, and that the company was looking into it1. Later, the same spokesperson told Ars the company was “fixing the issue with 2020 election queries in AI Mode,” attributing a technical gap that also affected 2022 Senate races.
Google’s public election policy pages, cited by Ars, say the company aims to surface official resources in results, including around the 2026 midterms1. Independent tests beyond Ars’s newsroom were not reviewed for this filing.
