With AI News Summaries, Verify Then Trust

Artificial intelligence (AI) assistants might be convenient, but they frequently provide problematic search results, especially where the news is concerned.

New research from the European Broadcasting Union (EBU) found that AI assistants routinely misrepresent news content, regardless of the language, territory, or AI platform in question.

The international study, News Integrity in AI Assistants, involved 22 public service media organizations (including the BBC, Radio France, CBC-Radio Canada, Czech Radio, and NPR) in 18 countries working in 14 languages. Professional journalists from the participating organizations evaluated more than 3,000 responses from AI tools ChatGPT, Copilot, Gemini, and Perplexity against key criteria, including accuracy, sourcing, distinguishing opinion from fact, and providing key context.

There were notable problems. Forty-five percent of all AI answers had at least one significant issue, and 20 percent contained major accuracy problems, including hallucinated details and outdated information. Some answers represented opinion or satirical content as facts. Looking at lower-severity issues, 81 percent of responses had at least one problem.

There has been improvement in recent months. A February 2025 BBC study found that 51 percent of AI assistant responses to BBC-specific news queries had significant issues, while only 37 percent did for the EBU study.

But sourcing is still a huge problem—31 percent of all responses studied for the EBU report had sourcing issues, including misattributing incorrect information to a news source, which can undermine the outlet’s perceived credibility. Gemini had a particularly high error rate for sourcing—72 percent of its responses had a significant sourcing issue, compared to below 25 percent for other AI assistant platforms.

AI assistants are already replacing search engines for many users, especially younger people; 15 percent of people under 25 years old use AI assistants to get their news, according to the Reuters Institute’s Digital News Report 2025. In separate research released this week, the BBC found that just 42 percent of UK adults say they completely trust AI to produce accurate summaries of information; that number rises to almost half when looking just at people under 35 years old.

While users are most likely to turn to AI to summarize complex news stories, those are the most difficult for AI to capture accurately, the EBU study found. Assistants struggled most with fast-moving stories with rapidly changing or updating information, intricate timelines involving multiple actors, detailed information, or topics that require clear distinction between facts and opinions and proper attribution of claims. Responses to fact-based and straightforward questions—such as the number of countries hosting the FIFA World Cup or Elon Musk’s birthplace—are less likely to contain significant issues.

The inaccuracies in AI assistant-produced news summaries can have significant effects on those users’ worldviews and their perception of the news, according to the EBU report.

It can also have ramifications for news publishers. Many users expect publishers and regulators to step in when things go wrong in AI summaries, and across all age groups, people held the source cited in the summary at least partly responsible for any errors, the BBC found.

“This research conclusively shows that these failings are not isolated incidents,” said EBU Media Director and Deputy Director General Jean Philip De Tender in a news 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.”