The BBC and the European Broadcasting Union have presented the largest study conducted to date on the behavior of artificial intelligence chatbots when asked about news. The results are alarming: Google's Gemini, OpenAI's ChatGPT, and Microsoft's Copilot provided incorrect or outright false answers in 45% of the analyzed cases. The study involved 22 public broadcasters from 18 countries, which conducted thousands of inquiries to these systems over several months. The method involved journalists from each country creating lists of verified news and then questioning the chatbots about those same topics, subsequently comparing the responses with the actual information. It was observed that errors particularly increased in matters of politics, economics, and complex social issues. Additionally, none of the chatbots warn the user when they are uncertain about an answer; instead, they generate fabricated data and present it with complete confidence. The differences between the systems also caught the researchers' attention. ChatGPT invented sources in 52% of cases, Gemini mixed different events in 48%, and Copilot altered relevant dates in 41%. A common problem is that all present false information with the same convincing tone they use for correct data, making it difficult for users to distinguish between truth and fiction without manual verification. The study provides concrete examples: when asked about recent elections in Europe, ChatGPT attributed non-existent statements to politicians, Gemini combined results from different countries, and Copilot modified the date of the elections. In another case, all three systems invented news headlines that never existed when responding about economic changes. The conclusion is clear: the heads of the 22 broadcasters warn that these chatbots pose a risk to democracy if used as a primary source of information. They recommend that users always verify important data with official sources before trusting the responses of artificial intelligence. Source: vandal.elespanol.com