What changed
Google AI Mode sometimes refuses to directly say that Joe Biden won the 2020 US presidential election. In tests reported on October 28, 2026, it instead directed users to official sources for the 2020 and 2024 results, while answering some older election questions plainly.
The behavior was inconsistent:
- Some users received direct answers to recent election questions.
- Repeating the same question sometimes produced a direct answer.
- Tests through Taiwan, Russia, France and Canada reportedly returned straightforward answers.
- Google Gemini, ChatGPT, Claude and America.gov’s AI model answered directly.
Why it matters
This is a small but revealing failure in a basic product promise: when someone asks an AI a factual question, they expect an answer, not a scavenger hunt.
Google AI Mode was able to provide accurate election information in some circumstances, then became evasive depending on the user, country, election cycle or wording. The cause is unknown, and the behavior’s scope and persistence remain unclear. But the practical problem is simple: a system can appear knowledgeable while quietly changing how willing it is to state politically sensitive facts.
That matters because AI search is increasingly used as a first stop for quick verification. Referring people to official records can be sensible when a question is ambiguous. Here, the question was direct, and the system sometimes answered comparable historical questions without hesitation. The inconsistency makes it harder to tell whether caution reflects a deliberate policy, a technical quirk or an unstable response pattern.
For now, the safest reading is also the least dramatic: Google AI Mode appears capable of answering these questions, but does not do so reliably in every context.
Comments
No comments yet.