Overview
- Users and reporters discovered Wednesday that Google Search’s built‑in AI feature, called AI Mode, often declined to say who won the 2020 presidential election and instead pointed people to official sources like the FEC or National Archives.
- Google’s standalone Gemini chatbot and other major models answered the same question directly, showing a product split where AI Mode applied stricter election filters than Gemini.
- Tests showed the behavior was location and interaction dependent: AI Mode usually returned a direct answer when accessed from non‑U.S. locations or after repeated questioning, which suggests the response depended on geolocation and query context.
- Google publicly acknowledged the problem, called it a “technical gap” in how the company verifies election answers using news/AP data and historic records, said some 2022 races were affected too, and reported that fixes are being deployed.
- The episode raises trust and safety concerns because Google’s AI overviews reach billions of users and any uneven guardrails could shape civic knowledge in the run‑up to the Nov. 3 midterm elections and beyond.