Major AI Chatbots Deliver Inaccurate News in Nearly Half of Responses, BBC-Led Study Reveals

Widespread Accuracy Problems in AI News Delivery

Four of the world’s most prominent AI chatbots routinely provide inaccurate or misleading news content to users, according to a landmark investigation conducted by the BBC on behalf of the European Broadcasting Union. The study, described as the largest of its kind, analyzed responses from OpenAI’s ChatGPT, Microsoft Copilot, Google Gemini, and Perplexity, finding that these systems misrepresented news content in almost half of all cases examined.

Widespread Accuracy Problems in AI News Delivery
Alarming Error Rates Across Platforms
Google Gemini Identified as Worst Performer
Concerning Examples of AI Misinformation
Systemic Issues in AI Design
Public Trust and Media Implications
Path Forward for AI Development

Alarming Error Rates Across Platforms

Researchers analyzed more than 3,000 responses from the AI assistants and discovered that 45 percent contained at least one significant issue, the report states. More concerning, 31 percent demonstrated serious sourcing problems, while one-fifth exhibited “major accuracy issues, including hallucinated details and outdated information.” When accounting for minor errors, a staggering 81 percent of responses included some form of mistake, sources indicate.

Google Gemini Identified as Worst Performer

Google’s Gemini displayed particularly troubling performance, with researchers identifying “significant issues” in 76 percent of responses it provided—double the error rate of other AI bots included in the study. Analysts suggest this poor performance stems primarily from Gemini’s sourcing problems, with significant inaccuracies appearing in 72 percent of responses. This error rate was three times higher than ChatGPT (24 percent), followed by Perplexity and Copilot (both 15 percent).

Concerning Examples of AI Misinformation

The investigation uncovered numerous instances where AI systems provided confidently stated but completely inaccurate information. Examples included ChatGPT incorrectly asserting that Pope Francis was still serving as pontiff weeks after his death, and Gemini denying that NASA astronauts had ever been stranded in space—despite two crew members having spent nine months stuck on the International Space Station. According to reports, Google’s AI bot responded to questioning by suggesting researchers were “confusing this with a sci-fi movie or news that discussed a potential scenario where astronauts could get into trouble.”

Systemic Issues in AI Design

The findings emerge shortly after OpenAI admitted that its models are programmed to sound confident even when uncertain, analysts suggest. In a September paper, the company conceded that AI bots are rewarded for guessing rather than admitting ignorance—a design flaw that encourages hallucinatory behavior. This tendency has manifested in embarrassing real-world scenarios, including lawyers representing Anthropic being forced to apologize to a US court after submitting filings containing fabricated citations invented by its Claude model.

Public Trust and Media Implications

An accompanying Ipsos survey of 2,000 UK adults revealed that 42 percent trust AI to deliver accurate news summaries, rising to half of respondents under 35. However, 84 percent indicated that a single factual error would significantly damage their trust in AI summaries, highlighting the substantial risks media outlets face from poorly performing algorithms. “This research conclusively shows that these failings are not isolated incidents,” said Jean Philip De Tender, EBU deputy director general. “When people don’t know what to trust, they end up trusting nothing at all, and that can deter democratic participation.”

Path Forward for AI Development

The report was accompanied by a toolkit designed to help developers and media organizations improve how chatbots handle news information and prevent them from bluffing when uncertain. As consumer use of AI chatbots continues to grow, the study underscores the urgent need for enhanced accuracy and transparency in AI-generated content, particularly in the critical domain of news and information dissemination.