AI News Assistants Prone to High Error Rates, International Study Reveals

Widespread Accuracy Problems in AI News Responses

Artificial intelligence assistants frequently deliver flawed news information, with significant errors appearing in 45% of responses, according to reports from a major international study. The European Broadcasting Union (EBU) research found that AI platforms “routinely misrepresent news content no matter which language, territory, or AI platform is tested.”

Widespread Accuracy Problems in AI News Responses
Systemic Issues Across AI Platforms
Performance Variations Between AI Models
Threats to Public Trust and Information Ecosystems
Industry Response and Improvement Efforts

The comprehensive evaluation involved 22 public service media organizations across 18 countries and 14 languages, analyzing approximately 3,000 news-related responses from leading AI chatbots. The study assessed OpenAI’s ChatGPT, Microsoft Copilot, Google Gemini, and Perplexity against critical criteria including accuracy, sourcing, and ability to distinguish opinion from fact.

Systemic Issues Across AI Platforms

Sources indicate that 81% of all responses contained at least minor problems, with sourcing emerging as the primary concern. The report states that 31% of answers showed serious sourcing issues involving missing, misleading, or incorrect attributions.

Accuracy problems closely followed, affecting 30% of responses with either hallucinated details or outdated information. In one notable example, ChatGPT reportedly claimed the current Pope was Pope Francis, who had supposedly died a month earlier and been succeeded by Pope Leo XIV. Another instance involved Copilot citing a 2006 BBC article as evidence for current bird flu vaccine trials., according to recent developments

Performance Variations Between AI Models

Analysts suggest Google Gemini demonstrated the poorest performance among tested models, with issues appearing in 76% of its responses—more than double the error rate of other platforms. Microsoft Copilot followed at 37%, with ChatGPT at 36% and Perplexity at 30%.

The research found that assistants particularly struggled with fast-moving stories, intricate timelines, and topics requiring clear fact-opinion distinctions. Nearly half of models reportedly had significant issues when responding to questions about whether former President Trump was starting a trade war.

Threats to Public Trust and Information Ecosystems

“This research conclusively shows that these failings are not isolated incidents,” EBU Media Director Jean Philip De Tender stated in a press release. “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 concerns come as AI assistants are rapidly becoming primary information sources for everyday users, potentially displacing traditional search engines. According to the Reuters Institute and University of Oxford, while using AI for news remains a minority activity, adoption has doubled since last year, with highest usage in Argentina and the United States among 18-24 year olds.

Industry Response and Improvement Efforts

Despite the concerning findings, the report indicates some platforms have shown improvement since similar research conducted earlier this year. Gemini demonstrated the most accuracy improvement, while ChatGPT and Perplexity showed no progress. However, sourcing issues remained problematic across platforms, with Gemini showing no improvement in this area.

AI companies continue expanding their news-focused offerings. OpenAI recently launched its web browser ChatGPT Atlas, while Google has integrated AI overviews into its search engine and announced comprehensive Gemini integration with Chrome. Perplexity’s AI-first browser Comet has faced security concerns after researchers demonstrated it could reveal user login information.

The researchers concluded that despite incremental improvements, “AI assistants are still not a reliable way to access and consume news,” maintaining their assessment from previous studies. With 48% of 18-24 year olds using AI to simplify complex stories, and 27% of adults over 55 doing the same, analysts suggest the reliability gap between user trust and actual performance represents a significant challenge for information ecosystems.

References

This article aggregates information from publicly available sources. All trademarks and copyrights belong to their respective owners.

Note: Featured image is for illustrative purposes only and does not represent any specific product, service, or entity mentioned in this article.

Broadcom has unveiled the Thor Ultra, an 800G Ethernet Network Interface Card designed for AI data centers. The card leverages PCIe Gen6 and new packet control technologies to address limitations of traditional RDMA, promising improved performance and vendor flexibility.

Broadcom’s New AI Networking Solution

Broadcom has introduced the Thor Ultra, described in reports as the industry’s first 800G AI Ethernet Network Interface Card (NIC). According to sources, the card is engineered for large-scale data centers that support advanced Artificial intelligence tools and workloads across hundreds of thousands of XPUs. The Thor Ultra’s implementation of PCIe Gen6 x16 is said to significantly boost throughput, addressing performance and scalability demands in AI-driven environments.