A landmark international study has revealed that popular artificial intelligence chatbots frequently misrepresent news and provide inaccurate information. The research, involving 22 public service media organizations, found significant issues in nearly half of the responses generated by leading AI assistants when questioned about current events.
The findings raise serious questions about the reliability of using AI tools like ChatGPT, Google's Gemini, Microsoft's Copilot, and Perplexity AI as primary news sources, a practice that is growing, particularly among younger audiences.
Key Takeaways
- A study by 22 public broadcasters found AI chatbots had significant issues in 45% of news-related answers.
- Major factual errors were present in 20% of the tested responses.
- Sourcing was a major problem, with 31% of answers having serious issues with attribution.
- Google's Gemini performed the worst on sourcing, with 72% of its responses showing significant issues.
- The study confirms that these failures are systemic and occur across multiple languages and countries.
A Widespread and Systemic Issue
An extensive investigation conducted by a consortium of 22 public broadcasters from 18 different countries has concluded that leading AI assistants systematically distort news content. The collaborative effort, coordinated by the European Broadcasting Union (EBU), evaluated 3,000 AI-generated responses to common news questions.
Journalists from organizations including the BBC and NPR tested four widely used chatbots: ChatGPT, Microsoft Copilot, Google Gemini, and Perplexity AI. The evaluation measured accuracy, sourcing, context, and the ability to distinguish fact from opinion. The results were consistent: 45% of all answers contained at least one significant issue.
Jean Philip De Tender, deputy director general of the EBU, emphasized the gravity of the findings. "This research conclusively shows that these failings are not isolated incidents," he stated. "They are systemic, cross-border, and multilingual, and we believe this endangers public trust."
"When people don't know what to trust, they end up trusting nothing at all, and that can deter democratic participation."
The study highlights a critical challenge as more people turn to AI for information. According to the Reuters Institute's 2025 Digital News Report, 7% of online news consumers already use AI chatbots for news, with that figure jumping to 15% for individuals under 25.
Breaking Down the Errors
The research uncovered specific and recurring types of failures across all tested platforms. While some issues were minor, a substantial portion involved serious inaccuracies and poor sourcing that could easily mislead a user.
Key Statistics from the Study
- Significant Issues: 45% of all answers had at least one major problem.
- Factual Errors: 20% of responses contained major factual inaccuracies.
- Sourcing Problems: 31% had serious issues with citing sources.
During testing, chatbots provided outdated information, such as naming a former German Chancellor or NATO secretary general as the current officeholder, long after leadership had changed. These are not nuanced errors but fundamental factual mistakes about major global figures.
While all four AI assistants showed problems, Google's Gemini had the most trouble with sourcing, with a staggering 72% of its responses demonstrating significant issues in this area. This means users are often unable to verify the information presented or understand its original context.
A Persistent Problem
This large-scale study builds on previous research, including a project by the BBC in February 2025 which found similar issues. At that time, more than half of the AI answers checked had significant problems, and nearly one-fifth of answers that cited BBC content introduced their own factual errors.
A comparison between the two studies shows only minor improvements over the past eight months, indicating that the core problems with AI-generated news summaries are not being resolved quickly.
"Despite some improvements, it's clear that there are still significant issues with these assistants," said Peter Archer, BBC program director of generative AI. "People must be able to trust what they read, watch and see."
The Impact on Public Trust
The proliferation of unreliable AI-generated news poses a direct threat to an informed public. When automated systems present falsehoods with the same confidence as facts, it erodes the foundation of trust in information.
The study's organizers argue that the way these AI models process and present information from trusted news sources is deeply flawed. They can strip away crucial context, misattribute facts, or invent information entirely—a phenomenon often referred to as "hallucination."
What is a "Hallucination" in AI?
In the context of artificial intelligence, a hallucination occurs when an AI model generates information that is nonsensical or factually incorrect but presents it as if it were true. This happens because the models are designed to predict the next most likely word, not to verify facts, leading them to create plausible-sounding but false statements.
This decontextualization of news is particularly dangerous. A factually correct piece of information, when presented without its surrounding context, can become misleading. The study found that AI assistants regularly fail to provide this necessary background, leaving users with a distorted view of events.
A Call for Accountability
In response to the findings, the media organizations behind the study are urging governments and AI companies to take immediate action. The EBU is pressing regulators in the European Union to enforce existing laws related to information integrity and media pluralism to hold AI developers accountable.
Furthermore, the EBU and other international media groups have launched a campaign called "Facts In: Facts Out." The campaign's core demand is simple: if an AI model is trained on factual news content, it must output factual information.
The campaign's organizers released a statement on the responsibility of AI companies. "When these systems distort, misattribute or 'decontextualize' trusted news, they undermine public trust," the statement reads. "AI tools must not compromise the integrity of the news they use."
As AI technology continues to integrate into daily life, the need for independent monitoring and robust regulation becomes more urgent. The study serves as a critical reminder that while AI offers powerful capabilities, its application as a reliable news source remains deeply problematic and requires significant oversight to protect the public's access to accurate information.