ModelSignal
Find a post
Actually Tested· Part 4

Stop asking AI for the news: 45% of answers had a serious problem.

45% of AI assistants' news answers had at least one significant issue, graded by journalists at 22 broadcasters.

13 Oct 2026 · Sources read 13 Oct 2026

Jump to the findings
The cover of the ModelSignal post: Stop asking AI for the news: 45% of answers had a serious problem.
The post’s cover as it appeared on social media.

We didn’t run this. EBU did. ModelSignal read the published work and checked every number below against it — the sources and the dates we read them are at the end.

22 newsrooms · 3,000 answers

45%

of AI news answers had

a serious problem.

EBU and BBC study, October 2025. 18 countries.

All four vs Gemini

ALL FOUR

All 4 assistants

45%

Answers with issues

Sourcing problems31%

GEMINI

Google Gemini

76%

Answers with issues

Main problemSourcing

Click the source.

Nearly half the answers had a serious issue. Gemini: three answers in four.

EBU + BBC · 22 broadcasters · 14 languages

SHARE OF 3,000+ ANSWERS

What went wrong

  • Any serious issue45%
  • Sourcing problems31%
  • Accuracy problems20%
  • Gemini, any issue76%

Graded by journalists at 22 public broadcasters, against their own standards.

The proof

The EBU and BBC release, October 2025. Read 13 Oct 2026.

Screenshot of the source: ebu.ch · news integrity report
ebu.ch · news integrity report

The full breakdown

Somebody else ran this study; we read it. Every number below is quoted from the EBU/BBC release and report page, with the date we read them.

The study

Coordinated by the European Broadcasting Union, led by the BBC, published 21 October 2025. Professional journalists at 22 public service broadcasters in 18 countries and 14 languages graded more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity for accuracy, sourcing, context and opinion vs fact.

Findings
Answers with at least one significant issue 45%
Serious sourcing problems 31%
Major accuracy problems (fabricated or outdated) 20%
Gemini, answers with significant issues 76% (the worst, mostly sourcing)

Why it matters

Reuters Institute Digital News Report 2025: 7% of online news consumers use AI assistants for news, 15% of under-25s.

Sources

  1. EBU — News Integrity in AI Assistants: an international PSM study (report page) Read 13 Oct 2026 · scope (22 PSM, 18 countries, 14 languages), 3,000+ responses, the four assistants, headline findings, publication date 21 Oct 2025
  2. EBU/BBC press release, republished by VRT: 'Largest study of its kind shows AI assistants misrepresent news content 45% of the time' Read 13 Oct 2026 · 45% / 31% / 20% / Gemini 76%, Reuters DNR usage figures, who led and coordinated, link to the BBC's February 2025 study

The catches

  • Questions were about news and current affairs; the grading criteria and graders belong to the broadcasters themselves.
  • Per-assistant figures other than Gemini's 76% are in the full report, not in the release we quote.
  • Assistants change weekly; the figures describe the answers collected for this study.
Every number, and where it’s quoted from
  • 45% of answers had at least one significant issue“45% of all AI answers had at least one significant issue”EBU/BBC release (VRT copy)
  • 31% sourcing problems“in 31% of cases, references were missing or misleading”EBU/BBC release
  • 20% accuracy problems“20% of responses contained major accuracy issues such as fabricated or outdated information”EBU/BBC release
  • Gemini worst, 76%“Gemini performed the worst, with significant issues in 76% of the tested responses”EBU/BBC release
  • 3,000+ responses, 22 broadcasters, 18 countries, 14 languages“22 public service media organizations across 18 countries, operating in 14 languages … Over 3,000 AI responses”EBU report page
  • four assistants: ChatGPT, Copilot, Gemini, Perplexity“ChatGPT, Copilot, Gemini, and Perplexity”EBU report page
  • graded by professional journalists against set criteria“Professional journalists assessed answers against criteria including accuracy, sourcing, distinguishing opinion from fact, and providing context”EBU report page
  • 7% of online news consumers use AI assistants for news; 15% of under-25s“7% of total online news consumers use AI assistants to get their news, rising to 15% of under-25s”EBU/BBC release citing Reuters Institute DNR 2025
  • built on the BBC's February 2025 study“built on research by the BBC published in February 2025”EBU/BBC release
  • published 21 October 2025“Report Publication Date: 21 October 2025”EBU report page
All posts