AI Assistants Get the News Wrong Almost Half the Time
A 2025 study by 22 public broadcasters found that 45 percent of AI assistants' answers about the news had at least one significant problem; months earlier Apple had paused AI news summaries after false headlines.
Topics: science and technology, society and culture
Example
What happened: In October 2025 a study led by the BBC and coordinated by the European Broadcasting Union, with 22 public broadcasters in 18 countries, had journalists check more than 3,000 answers about the news from ChatGPT, Copilot, Gemini and Perplexity. 45 percent had at least one significant issue; 31 percent had serious sourcing problems and 20 percent major accuracy errors, such as invented or outdated details. Earlier, in January 2025, Apple paused AI summaries of news notifications after one falsely told BBC readers that a murder suspect had shot himself.
Use it for: AI tools that summarise the news can spread errors under the name of trusted news outlets, so faster access to news does not mean more accurate news.
Why it matters: an AI summary looks like a neutral digest of the source, so readers blame or trust the outlet named beside it; when the summary is wrong, the error carries the outlet's credibility with it.
Limit: The study tested free versions of four assistants over a set period, and AI tools change quickly, so error rates may since have fallen; it measures answers, not whether readers believed them.
Key facts
- A study coordinated by the EBU and led by the BBC, with 22 public service media organisations in 18 countries and 14 languages, evaluated more than 3,000 news answers from ChatGPT, Copilot, Gemini and Perplexity.
- 45 percent of the AI answers had at least one significant issue, 31 percent had serious sourcing problems and 20 percent had major accuracy issues; Gemini had significant issues in 76 percent of its responses.
- In January 2025 Apple paused AI-generated notification summaries for news apps after false alerts, including one that wrongly told BBC readers that murder suspect Luigi Mangione had shot himself.
How to use this example in a GP essay
Will artificial intelligence make people better or worse informed?
Claim
AI tools that summarise the news can spread errors under the name of trusted news outlets, so faster access to news does not mean more accurate news
How the evidence supports it
an AI summary looks like a neutral digest of the source, so readers blame or trust the outlet named beside it; when the summary is wrong, the error carries the outlet's credibility with it
Limitation
The study tested free versions of four assistants over a set period, and AI tools change quickly, so error rates may since have fallen; it measures answers, not whether readers believed them.
Relevance
It gives a large, recent, multi-country measurement of AI accuracy on news, and a concrete company response, for media and technology questions.
Limitations
- The study tested free versions of four assistants over a set period, and AI tools change quickly, so error rates may since have fallen; it measures answers, not whether readers believed them.
Evaluations
science and technology evaluation
Support
Nearly half of answers with a significant problem, across four leading tools and 14 languages, shows the errors are built into how current assistants handle news, not a quirk of one product.
Counterargument
AI models improve quickly, and the tools tested in 2025 may already have been replaced by better versions, so the figure may overstate the problem today.
Rebuttal
Improvement should be shown by similar independent tests, not assumed; until then, a 45 percent error rate is the best evidence available.
Additional support
Apple's decision to pause its feature, rather than defend it, shows a major technology firm accepting that its AI was not ready to summarise news.
society and culture evaluation
Support
Readers saw a false Apple summary under the BBC's name, so AI errors damaged trust in a news outlet that had not made the mistake.
Counterargument
Human journalists also make errors, and many readers skim headlines anyway, so AI summaries may not be much worse than existing news habits.
Rebuttal
Human errors are usually corrected and traced to an author; AI errors are produced at scale, change from one answer to the next and are rarely corrected publicly.
Additional support
The study found sourcing was the biggest problem, which matters because readers cannot check a claim if the assistant names the wrong source or none.
Sources
- European Broadcasting Union (2025-10-22): AI's systemic distortion of news is consistent across languages and territories: international study by public service broadcasters
- TechCrunch (2025-01-16): Apple pauses AI notification summaries for news after generating false alerts