International research: AI assistants often distort news content, with 45% of answers containing significant questions

A recent study coordinated by the European Broadcasting Union (EBU) and led by the British Broadcasting Corporation (BBC) has found that artificial intelligence assistants, which serve as daily information sources for millions of people, often distort news content when testing any language, region, or AI platform. This unprecedented, wide-ranging, and largest international study was released at the EBU Press Conference in Naples. 22 public media organizations (PSMs) from 18 countries and using 14 languages participated in the study, revealing multiple systemic issues present in four mainstream AI tools. According to IT Home, professional journalists from various participating public media evaluated over 3000 responses generated by ChatGPT, Copilot, Gemini, and Perplexity based on key criteria such as accuracy, source labeling, distinguishing facts from viewpoints, and providing background information. The main research findings are: 45% of all AI responses have at least one significant issue. 31% of the answers have serious information source issues, including missing, misleading, or incorrectly citing sources. 20% of the answers have significant accuracy issues, including fabricated details and outdated information. Gemini performed the worst, with 76% of its answers having significant issues, more than twice that of other AI assistants, mainly attributed to its poor information traceability ability. Compared with the research results of BBC earlier this year, some AI tools have improved, but the error rate is still at a high level. Artificial intelligence assistants have gradually replaced traditional search engines and become the preferred information entry point for many users. According to the Reuters News Institute's 2025 Digital News Report, 7% of online news consumers worldwide obtain news through AI assistants, and this proportion is as high as 15% among people under the age of 25. Jean Philip De Tender, Director and Deputy Director General of EBU Media, said, "This study clearly shows that these issues are not isolated events, but rather have systemic, transnational, and multilingual characteristics. We believe that this is jeopardizing the public's trust in the media. When people are unable to judge what is trustworthy, they may eventually lose faith in anything, which will weaken the participatory foundation of democratic societies Peter Archer, Director of the BBC's Generative Artificial Intelligence Program, said: 'We have high expectations for AI and believe it can help us create more value for our audience.'. But the prerequisite is that people must be able to trust the content they read, watch, and come into contact with. Although there have been some improvements, these AI assistants still have significant issues. We hope these technologies can succeed and are willing to collaborate with AI companies to create positive impacts for audiences and society. ” The research team has also released the 'News Integrity in AI Assistants Toolkit', aimed at providing practical solutions to the problems revealed in the report. This tool includes two aspects: improving the response quality of AI assistants and enhancing user media literacy. Based on the large number of cases and insights collected in this study, the toolkit focuses on two core questions: "What kind of AI assistant response is considered high-quality? ”And 'Which issues urgently need to be fixed?' In addition, the EBU and its member institutions are urging EU and national regulatory agencies to strictly enforce existing laws on information authenticity, digital service regulations, and media diversity. Given the rapid development of AI technology, continuous independent monitoring is crucial. To this end, EBU is exploring the establishment of a normalized, rolling research mechanism to track the performance of AI assistants over the long term. This study is based on preliminary research released by BBC in February 2025, which first revealed serious flaws in AI's ability to process news content. This second phase of research will expand its scope to the global level, further confirming that such issues are universal and not limited to a specific language, market, or AI assistant. According to a supplementary study released by the BBC on the same day, the public's habits and perceptions of using AI assistants for news are also worrying: currently, more than one-third of British adults believe that AI generated news summaries are accurate, and among people under 35 years old, this proportion is nearly half. These findings have raised significant concerns: many people mistakenly believe that AI generated news summaries are accurate and reliable, when in fact they are not; When they discover errors, they often blame both news agencies and AI developers - even if the errors are entirely caused by AI assistants. Over time, such issues may seriously damage the public's trust in the news itself and its brand.