AI is getting better at election facts, but voters shouldn’t rely on it

New research from States United Democracy Center tested ChatGPT and Google Search's AI interface on basic election questions. The article reports that the AI systems have improved in their responses to these questions. However, the research found that the systems are nowhere near reliable enough to serve as a main source of accurate or complete information for voters. The article's title summarises this dual finding: AI is getting better at election facts, but voters should not rely on it.

According to the article, the testing focused on basic election questions, though it does not specify the exact questions or the scoring criteria used. Both ChatGPT and Google Search's AI interface were included in the research. The article states that both showed improvement. It also states that the improvement is not enough to make them trustworthy as a primary information source. The article does not provide specific examples of wrong answers or measure the rate of inaccuracy.

The article does not provide specific statistics or company responses. It presents the research as evidence that AI election information remains unreliable. It also suggests that voters should continue to rely on official sources for election information. The article notes that progress has been made but characterises the current state as insufficient for voters to depend on. The headline warns against relying on AI despite the reported improvements.