AI Chatbots Frequently Misrepresent News, Study Finds
A new study reveals that leading artificial intelligence models, including ChatGPT, Gemini, Copilot, and perplexity, routinely provide inaccurate information when questioned about current events, with nearly half of all responses containing important errors.Teh research, published Wednesday by the European Broadcasting Union (EBU) and the BBC, assessed over 2,700 responses from the AI assistants.
Twenty-two public media outlets from 18 countries and representing 14 languages participated in the study, posing a standardized set of questions to the AI models between late May and early June. 45 percent of the responses exhibited at least one “significant” issue. Sourcing was the most prevalent problem, appearing in 31 percent of responses, followed by factual inaccuracies (20 percent) and a lack of contextual awareness (14 percent).
The study highlighted specific errors, such as Perplexity incorrectly stating the legality of surrogacy in Czechia and ChatGPT erroneously identifying Pope Francis as the current pontiff after his reported death. Gemini demonstrated the highest rate of significant issues, with 76 percent of its responses affected, primarily due to sourcing problems.
“They have not prioritized this issue and must do so now,” stated Jean Philip De Tender, EBU’s deputy general, and pete Archer, the head of AI at the BBC, in the report’s foreword. They also urged tech companies to increase transparency by regularly publishing results broken down by language and market.
Requests for comment from OpenAI, Google, Microsoft, and Perplexity were not immediately returned.
