**A major international study led by the BBC and coordinated by the European Broadcasting Union has found that AI assistants distort or misrepresent news content in 45% of responses, raising deep concerns about the reliability of AI-generated information and its growing influence on how people consume news.**

![BBC Study Reveals AI Assistants Get Nearly Half of News Facts Wrong
](https://www.stanventures.com/news/wp-content/uploads/2025/10/ChatGPT-Image-Oct-22-2025-06_34_21-PM.png)

It’s one thing for AI to suggest dinner recipes or summarize a movie plot. But when it comes to news, the cost of error is far greater. 

That’s exactly what a new [BBC-led study](https://www.bbc.co.uk/aboutthebbc/documents/audience-use-and-perceptions-of-ai-assistants-for-news.pdf) has found. 

The investigation, coordinated by the European Broadcasting Union (EBU), reveals that some of the most popular AI assistants, including ChatGPT, Copilot, Gemini, and Perplexity, frequently distort or even invent information when asked about news and current events.

Researchers tested these tools across 18 countries and 14 languages in what they call the largest and most detailed study of its kind. 

The findings, presented at the EBU News Assembly in Naples, paint a stark picture of [how unreliable AI has become](https://www.stanventures.com/news/hallucination-rates-spike-in-openais-o3-o4-mini-models-2490/) as a gateway to information.

## A 45% Failure Rate

Out of more than 3,000 AI-generated answers reviewed by professional journalists, 45% contained at least one major error. 

Some confused facts with opinions, some provided misleading sources, and others confidently stated things that simply weren’t true.

Even more concerning, 31% of the responses had serious sourcing problems, either missing attributions, linking to unrelated material, or citing sources that didn’t exist. 

And 20% contained outright factual mistakes, including hallucinated quotes and outdated claims.

Among the platforms tested, Google’s Gemini performed worst, with 76% of its news answers showing major issues, largely due to poor sourcing.

Although the BBC’s earlier report in February 2025 showed some improvement since then, the pattern shows that AI assistants still struggle with the fundamentals of responsible news communication.

## Trust Without Verification

Despite these flaws, the study reveals that [public confidence in AI](https://www.stanventures.com/news/who-do-you-trust-online-americans-are-more-skeptical-than-ever-2116/) remains surprisingly high. 

According to the BBC’s accompanying audience research, over a third of UK adults trust AI assistants to deliver accurate news summaries. Among those under 35, that trust rises to nearly half.

This misplaced confidence is one of the most serious implications of the findings. 

When users see AI-generated news summaries, they often assume the information has been verified, especially when the tone is authoritative. But when mistakes occur, blame often falls not only on the AI company but also on the original news publisher whose content may have been misrepresented.

## Why This Matters More Than Ever

The danger isn’t only about getting a few details wrong. The larger issue is how AI reshapes how people discover and believe news. 

As AI assistants increasingly replace traditional search engines, they’re becoming the first and sometimes the only source of news for millions.

According to the [Reuters Institute’s Digital News Report 2025](https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2025/dnr-executive-summary), 7% of all online news consumers already use AI assistants as their main gateway to news, and that figure jumps to 15% among people under 25.

This means that if AI systems consistently misinterpret or editorialize information, public understanding of world events could become skewed at scale. 

## Fixing the Flaws Before It’s Too Late

To address the findings, the EBU and BBC have launched a [News Integrity in AI Assistants Toolkit](https://www.bbc.co.uk/aboutthebbc/documents/news-integrity-in-ai-assistants-toolkit.pdf%20), a practical guide for both AI developers and media professionals. It highlights what makes a good AI-generated response, where the failures lie, and how both sides can collaborate to improve.

The report calls for several urgent steps:

- AI companies should publish transparent data on error rates by language and country.
- News publishers must demand clearer attribution when their content is used.
- Regulators should consider intervention if voluntary measures fall short.
- And audiences need better education about how AI works and where its limits are.

The BBC and EBU are also advocating for continuous, independent monitoring of AI assistants, recognizing that the technology changes faster than public understanding.

## Beyond Blame: A Call for Collaboration

Despite the harsh findings, the tone of the report isn’t anti-AI. 

Both the BBC and EBU acknowledge the potential of artificial intelligence to make journalism more accessible and engaging. But they insist that innovation must not outpace accountability.

The goal, they argue, is to make sure truth survives the upgrade. That means AI developers, journalists, and policymakers need to work together, not against each other.

## What Readers Can Do Right Now

Until AI assistants become more reliable, users can take a few simple steps to protect themselves from misinformation:

1. **Double-check the source.** If an AI gives you a summary, look up the original article.
2. **Be skeptical of confidence.** The smoother the tone, the easier it is to miss mistakes.
3. **Don’t rely on one answer.** Ask multiple tools, or better yet, read multiple outlets.
4. **Learn how AI generates text.** It doesn’t think; it predicts. That difference matters.
5. **Support real journalism.** Credible reporting still depends on human judgment, not algorithms.

## Key Takeaways

- 45% of AI-generated news responses contain serious factual or sourcing errors.
- Younger users are the most likely to trust AI news summaries, despite their flaws.
- Sourcing errors are the biggest weakness, with 31% of answers citing unverifiable material.
- Publishers risk reputational damage when AI assistants distort their content.
Experts call for collaboration and transparency to restore trust before misinformation spreads further.