The European radio and tv union (EBU) has conducted a survey of four AI services (ChatGPT, Copilot, Gemini, and Perplexity) and how they respond to news. The test was carried out in 18 countries in 14 different languages and shows that in 45 percent of the cases, the AI services' responses contained something significantly incorrect.
The different services were varying in their ability, or disability depending on how you look at it, when it comes to things like source references, neutrality, and relevance.
Among the less positive figures in the report, Google's Gemini stands out, where the service in almost three out of four responses could not provide correct sources for its claims.
However, the accuracy of AI is improving rapidly. In a BBC survey from the beginning of the year, every other response contained inaccuracies – a direct comparison with the survey shows that the figure is now down to just over a third.
At the same time, AI is taking over more and more of the content on the internet. According to a report from the search optimization company Graphite, the majority of articles, including news articles and other articles, published on the internet are now made by AI.
