A new study coordinated by the European Broadcasting Union (EBU) and led by the BBC has found that AI assistants — now used by millions of people as daily news gateways — routinely provide incorrect or misleading information. The worst performer was Gemini.
Global research reveals systemic flaws
Launched at the EBU News Assembly in Naples, the project involved 22 public-service media organizations from 18 countries working in 14 languages.
Professional journalists evaluated over 3,000 answers from ChatGPT, Copilot, Gemini, and Perplexity according to core journalistic standards such as factual accuracy, sourcing, separation of opinion and fact, and contextual clarity.
Key findings:
- 45 % of responses contained at least one major problem.
- 31 % showed serious sourcing issues, including missing or misleading attributions.
- 20 % had factual errors or outdated information.
- Gemini performed worst, with significant issues in 76 % of its answers — more than twice that of other systems, largely due to weak sourcing.
A comparison with earlier BBC research shows some improvement but persistently high error rates.
Younger audiences increasingly turn to AI for news
According to the Reuters Institute Digital News Report 2025, 7 % of online news consumers already rely on AI assistants for information — a figure rising to 15 % among those under 25.
EBU Media Director Jean Philip De Tender stated:
“These shortcomings are not isolated incidents but systematic, cross-border and multilingual. When people stop knowing what to trust, they eventually trust nothing — and that undermines democracy.”
Peter Archer, Director of the BBC Generative AI Program, added:
“We’re excited about AI’s potential, but audiences must be able to trust what they read, watch and hear. Despite some progress, major issues remain.”
Toolkit launched to strengthen news integrity
The research team released an AI Assistant News Integrity Toolkit designed to improve both response quality and media literacy among users. It addresses two core questions:
- What does a reliable AI response to a news query look like?
- Which common problems must be fixed?
The EBU and its members are urging the EU and national regulators to enforce existing laws on information integrity, digital services, and media pluralism. Given the fast-moving nature of AI, they emphasize the need for independent, ongoing monitoring of AI assistants.
The findings build on a February 2025 BBC study that first highlighted AI’s weaknesses in news processing and now confirm that these issues are systemic — independent of language, market or model.
A separate BBC survey found that many users assume AI news summaries are accurate. Just over one-third of UK adults say they trust AI-generated summaries, but confidence drops sharply among those under 35.
Researchers warn that such inaccuracies can erode public trust in journalism and news brands over time.
