Study: 45% of AI news responses contain serious errors

A recent study reveals that nearly half of all AI-generated news responses contain serious errors, raising urgent questions about our growing reliance on artificial intelligence for information. As AI assistants become the default way many people consume news—embedded in everything from search engines to smart devices—these accuracy problems pose significant risks for both individual decision-making and public discourse.

The European Broadcasting Union (EBU), a consortium of public broadcasters across Europe, tested more than 3,000 AI-generated responses across 14 languages using popular assistants including ChatGPT, Google Gemini, Microsoft Copilot, Claude, and Perplexity. Their findings paint a troubling picture of AI reliability when it comes to current events and factual information.

The scale of AI’s accuracy problem

The EBU study uncovered systematic issues across all major AI platforms. Forty-five percent of responses contained at least one significant error, while 81% had some form of problem ranging from outdated information to misleading phrasing. Perhaps most concerning, 31% were flagged for sourcing problems—including fabricated references, missing citations, or incorrectly attributed sources.

Twenty percent contained major factual inaccuracies, such as misreporting current events or incorrectly attributing quotes to public figures. While the study didn’t publicly rank individual assistants, internal figures reportedly showed that Google Gemini struggled particularly with sourcing accuracy, while ChatGPT and Claude showed inconsistent performance depending on which version users accessed.

These error rates are significantly higher than previous estimates. Earlier research suggested ChatGPT was wrong about 25% of the time, but the new data reveals the problem may be nearly twice as severe when specifically examining news-related queries.

Why AI news errors matter more than ever

The timing of these findings is particularly significant given AI’s rapid integration into daily information consumption. According to the Reuters Institute for the Study of Journalism, 15% of Generation Z users already rely on chatbots as a primary news source. Meanwhile, AI-powered features like Google’s AI Overviews and ChatGPT’s real-time search capabilities are becoming default options for millions of users seeking quick answers.

This shift creates a perfect storm for misinformation spread. Unlike traditional news sources, AI assistants often don’t clearly surface their sources or distinguish between established facts and speculation. When an AI confidently summarizes breaking news but omits publication details, timestamps, or opposing viewpoints, users may unknowingly absorb incomplete or outdated information with a false sense of confidence in its accuracy.

The implications extend beyond individual users. If millions of people daily consume flawed or biased AI summaries, it could systematically distort public understanding of important issues. Traditional news outlets face a double burden: losing traffic to AI interfaces while seeing their original reporting potentially misrepresented or stripped of crucial context.

Testing AI assistants on real news queries

To understand these problems in practice, I tested three leading AI assistants—ChatGPT, Claude, and Google Gemini—with the same current events question: “What’s the latest on the US debt ceiling deal?”

Claude provided the most accurate and useful response, correctly identifying the timeframe of recent debt ceiling negotiations and placing them in proper historical context. It accurately explained that the debt ceiling was reinstated in January 2025 following the suspension under the Fiscal Responsibility Act of 2023, and that new negotiations were needed to prevent a potential default. The response delivered core information clearly without speculation or unnecessary complexity.

ChatGPT’s response contained a critical flaw: it cited news articles with future dates, including references to “Today” and specific dates in 2025 that hadn’t occurred yet. While the background information was generally useful, these fictional citations severely undermined credibility. The response was well-structured but buried the actual answer to the question under generalized concerns and future projections.

Gemini correctly identified recent developments and provided solid context, but concluded by introducing an unrelated government shutdown issue without clearly explaining its connection to debt ceiling negotiations, potentially confusing users about distinct policy issues.

Protecting yourself when using AI for news

Given these accuracy challenges, users who choose to rely on AI for news consumption should adopt more strategic approaches to their queries and verification processes.

Rather than asking broad questions like “What’s happening in the world?” or “Tell me about recent news,” frame requests more specifically. Ask for sources upfront by adding phrases like “Give me links to recent, credible news outlets covering this topic.” Time-stamp your queries with phrases such as “As of today’s date, what’s the latest development in…”

Cross-checking information across multiple AI assistants can reveal discrepancies that signal potential accuracy issues. When responses vary significantly between platforms, that’s often a red flag warranting additional verification through traditional news sources.

Don’t stop at AI summaries for important information. If something sounds surprising or consequential, ask for links to full articles and review the original reporting. Treat AI assistants as starting points for research rather than authoritative sources, using them to surface relevant headlines while verifying facts through established journalism outlets.

Most importantly, maintain healthy skepticism about AI-generated news summaries, especially for breaking news or complex policy issues where context and nuance matter significantly.

The broader implications for information reliability

The EBU findings highlight a fundamental tension in how we consume information in the AI age. While these tools offer unprecedented convenience for quick research and news updates, their accuracy limitations create new risks for both individual decision-making and collective understanding of important issues.

For businesses increasingly integrating AI into research and decision-making processes, these findings underscore the importance of verification protocols. Companies relying on AI assistants for market research, competitive intelligence, or regulatory updates should implement systematic fact-checking procedures to avoid decisions based on flawed information.

The solution requires action from multiple stakeholders. AI companies need greater transparency in their sourcing systems and more robust accuracy safeguards. Users need better education about AI limitations and verification techniques. News organizations need clearer strategies for maintaining relevance and authority in an AI-mediated information landscape.

Until AI assistants can consistently cite sources accurately, distinguish between established facts and speculation, and update information in real-time, every AI-generated response warrants careful scrutiny. For breaking news and critical decisions, the most reliable approach may still be the traditional one: going directly to established news sources rather than relying on AI intermediaries.