Source: The Guardian

Demos ran a simulated inquiry ahead of the Holyrood election in May 2026 and found AI services supplied misinformation in 34% of 75 questions posed across three real constituencies. The simulation tested five free AI tools — including ChatGPT, Google Gemini, Replika and Grok — and reported that the systems sometimes invented fictitious scandals and candidates, provided the wrong date for the election, wrongly claimed voters needed ID in Scottish polling stations, and misassigned candidates to contests.

Demos’ report, Electoral Hallucinations, gives specific error rates: Replika 56% errors, ChatGPT 46%, Google Gemini 22% and Grok 9%; Google AI Overviews answered only 11% of prompts and was disregarded. The thinktank also found nearly half of responses lacked official sources or external links, links were sometimes broken, and ChatGPT citations were at least a year out of date 44% of the time. A poll of 2,005 British adults commissioned by Demos found 20% had used AI chatbots or search tools for information on parliamentary elections in Scotland and Wales and for English local councils, which Demos equates to about 10 million people across the UK. The Electoral Commission and Demos urged ministers to introduce clearer duties on AI platforms, statutory safeguards on accuracy, liability under UK defamation and electoral law, and requirements allowing independent testing of models and training data; the Department for Science, Innovation and Technology said defending elections was “an absolute priority.” Replika said its chatbot is not designed for fact‑checking and informs users of that purpose, while OpenAI said Demos appeared to be using an out‑of‑date version and noted users can instruct ChatGPT to search the web.