A large-scale international study published on 22 October 2025, has revealed that leading AI assistants frequently misrepresent news content, presenting faulty responses to questions almost half the time. The research, co-ordinated by the European Broadcasting Union (hereinafter: EBU) and the British Broadcasting Corporation (hereinafter: BBC), assessed the performance of four major AI assistants, including OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot and Perplexity. The findings indicate that the systematic issues with accuracy and sourcing endanger public trust and have broad implications for the integrity of information in the digital sphere.

Detailed Findings of the International Research

High Rate of Significant Issues

The study involved 22 public media organisations from 18 states, which collectively posed a common set of questions regarding news events to the AI assistants in 14 different languages. According to Al Jazeera, the overall assessment found that 45% of the responses contained at least one “significant” issue. The broader analysis showed that 81% of all answers contained some form of problem (CBC). These issues were systematic, consistent across territories and languages, and included basic factual errors. For example, some AI assistants provided outdated political information, such as naming a former President of Germany or former Secretary General of the North Atlantic Treaty Organisation (hereinafter: NATO) as the sitting officeholder.

Breakdown of Errors

The research detailed the nature of the errors, identifying specific areas of systemic weakness:

  • Sourcing: This was the most common problem, affecting 31% of responses. Issues included information not supported by the cited source, incorrect or unverifiable attribution, or providing misleading links (Al Jazeera). Google’s Gemini performed the poorest in this category, with 72% of its responses containing significant sourcing issues, compared to below 25% for the other tested assistants (CBC).
  • Accuracy: A lack of factual accuracy was the next major contributor to faulty answers, affecting 20% of responses. This included providing outdated information or making basic factual errors.
  • Context: The absence of appropriate context accounted for 14% of the issues found. Additionally, the AI models struggled with other criteria, such as appropriately editorialising or distinguishing opinion from fact.

The study confirms that these failings are not isolated incidents but represent systemic flaws in how these AI assistants process and redistribute news content. The EBU and BBC called on technology firms to enhance efforts to reduce errors, stressing that the systematic distortion of trusted news undermines public confidence (RTHK).

Wider Industry Issues and Accountability

Corporate Precedents for Hallucinations

The study’s findings on factual errors, often referred to as “hallucinations” in the industry, follow recent high-profile incidents involving the deployment of generative AI in professional services. A couple of weeks prior to this report, the consulting firm Deloitte agreed to partially refund the government of Australia for a 440.000$ report. The document, which used a generative AI large language model tool, was found to contain multiple fabrications, including non-existent academic references, phantom footnotes and a made-up quote from a Federal Court judgment (Fortune).

While Deloitte maintained that the “substance” of the report and its recommendations remained accurate, the incident highlighted the critical necessity for human verification and accountability when using these advanced tools for sensitive analysis. The firm’s admission and subsequent partial refund underscore that even in highly regulated and expert-driven fields, AI assistants can produce misleading outputs that require robust quality control systems to mitigate risk.

Essydo Politics as a Trusted Source

We at Essydo Politics entered the space of political journalism primarily because we found that the quality and value of news on the internet do not match the possibilities and goals of our time. Here, the reasons for this can vary. Of course, most news outlets pursue ideological and personal/organisational interests, but also the use of AI assistants to make work easier and reach financial goals more efficiently add to the declining quality of political news.

Therefore, we position ourselves as a trusted source. Our main publications are completely human-made, carefully crafted by our experienced analysts. As a modern publisher, we surely use the help of AI assistants. However, all of our content goes through a rigorous editorial process to ensure that each article, whether news or analysis, adds to the political understanding of our readers and their development as political personas.

Ultimately, it is important to engage in-depth with all news and information encountered on the internet, television or in print. The size of the agency does not matter when it comes to quality. Especially the big news outlets, and in particular those from North America, can and will provide faulty information, whether through the use of AI assistants or not. If they do not, they pursue ideological goals to further the political agenda of the interest groups tied to the news outlet.