Flutterby™! : AI lying about elections

A Scottish Post: The New Election Threat: Disinformation Inside the Answer. Via.

Looks like the source of this is from: Demos: Electoral Hallucinations: Safeguarding UK elections in the world of LLMs and AI chatbots

Our snapshot testing on March 27th 2026 found that - across ChatGPT, Gemini, Google AI Overviews, Grok, and Replika - just over a third (34.1%) of responses to questions about the Scottish elections contained factual errors (109 of 320 total responses). To break down these inaccurate responses, 8.75% (28 of 320) were entirely inaccurate and 25.3% were partly accurate but with errors (81 of 320). Partly factual responses could be particularly misleading as their errors were sometimes much harder to spot.